Something is hilariously off here: Why should I pay $10 and be forced to use it by the end of the month, while I can pay $10 and have it last as long as I want?
How would that be? They are already charging as much as the underlying providers. They can hardly expect to have any customers if they are charging more.
Enterprise sales will be the answer. Microsoft will have some story that convinces an exec eight levels up the org chart from the normal users that this is an essential product they need to overpay for. Given their existing relationshipsand immense sales team they'll probably have success.
That story is data governance. Corporate already have a data-agreement with MS, storing all their data there. Github copilot is covered by that, while a individual agreement with e.g. anthropic needs lawers involved.
It’s precisely this, and, to be fair, it’s a rational approach given a Data Security Exhibit starts at 6 weeks and can hit 6 months to complete. That being said, I work with regulated data, so YMMV.
Microsoft is simply the default answer for most large corporations. Getting access to some Microsoft subscription is very easy, because of the existing framework agreements, Microsoft providing any and all compliance slopuments needed and already being pre-cleared for corporate data etc. Meanwhile trying to use another provider (e.g. Anthropic) would be a one year endeavor, minimum.
e.g. if on an annual plan? 0x will be gone, but there are okay 1x and 0.3x models left. I am pretty much curious how the early may test invoicing will look like. current setup of tools etc. is way too chatty eats up 1+M token per PRU easily. not sure how much is cached.
Can you elaborate please? I still clone my private repos and work on them using OpenRouter Credits and OpenCode (or Copilot itself: It supports OpenRouter BYOK). No?
I have to wonder if it's because of how many Enterprise customers they have who have standardized on Github Copilot and gotten it through the gauntlet of legal approvals etc.
This rings true to me as someone who's worked at a few large corps like this. A price hike does not change things when there is a mandate to use MS products over other vendors.
I'm wondering if they're basically saying they're going to give $10/month free API credits to students and open source maintainers and so on... while otherwise getting out of the consumer portion of this space.
they're downsizing free github copilot pro for open source maintainers. At the very least, it looks like small open source projects got their free copilot pro cut off
You can pool credits through open router (afaik, I'm only using a single user account), but if you top-up $10 per user, per month, any unused credits will rollover.
Tbh I think it still works, but only because the new allowance will likely get used very quickly within a billing cycle - I'm expecting this change to increase our orgs bill significantly based on how many API credits with open router I consume in a weekend using a single agent in a pairing style.
The pooling will only be useful if you have a bunch of infrequent/low usage users that you still want to have licenses.
Lots of us have noticed that usage limits for Claude have been nerfed in recent weeks/months.
If anything, these new multipliers are more transparent than anything OpenAI or Anthropic have communicated regarding actual costs and give us a more realistic understanding of what it's costing these providers.
The fact that we were able to get such a substantial amount of usage for $20/$100/$200 a month was never meant to last and to think otherwise was perhaps a bit naive.
This feels like a strategy from the ZIRP era of tech growth where companies burned investor capital and gave away their products and services for free (or subsidized them heavily) in order to prioritize user acquisition initially. Then once they'd gained enough traction and stickiness they'd then implement a monetization strategy to capitalize on said user base.
However, inference costs for entirely good enough models are likely to keep declining in the future. We're probably hitting diminishing returns on model size and training. The new generations aren't quantum leaps anymore, and newer generations of open source models like DeepSeek are likely to start getting good enough.
There's going to be a limit to how much they can raise prices, because someone can always build out a datacenter and fill it up with open source DeepSeek inference and undercut your prices by 10x while still making a very good ROI--and that's a business model right there. Right now I'm sure there's a lot of people who will protest that they couldn't do their jobs with lesser models, but as time goes on that will get less and less. Already right now the consumers who are using AI for writing presentations, cooking recipe generation and ELI5 answers for common things, aren't going to be missing much from a lesser model. That'll actually only start to get cheaper over time.
Also for business needs, as AI inference costs escalate there comes a point where businesses rediscover human intelligence again, and start hiring/training people to do more work to use lesser models--if that is more productive in the end than shelling out large amounts of cash for inference on the latest models. [Although given how much companies waste on AWS, there's a lot of tolerance for overspending in corporations...]
> because someone can always build out a datacenter and fill it up with open source DeepSeek inference and undercut your prices by 10x while still making a very good ROI-
Not sure how it all works out. Currently trillion dollar companies can't make a native app for platforms. Everything is just JS/Electron because economics does not work for them.
And here companies can make GW data center running very expensive GPUs for 1/10th of current prices. Sound little fanciful to me.
The price you pay for anthropic must include the price of training new and better models which is incredibly costly. If you use the models someone else already spend money to develop you don’t need to pay this price.
And at some point even frontier model costs will hopefully come down (if there is still a meaningful difference between closed and open source models at that point) as all of the compute that's being built out right now comes online.
Yups... Mythos is the smallest possible leap. Not a standard model generation advance, not even a version point advance. Just the smallest possible quanta of a change. We are absolutely hitting a plateau any day now. Any day. Any time. Any second now. Yup. Right now! Surely!
Yeah. AI progress is insanely fast if you compare it to anything else. Where else is a one year old technology already hopelessly outdated? 10 years ago is basically stone age.
I am continually tripped out by the fact when I was 16, I didn't have a 'smartphone' beyond a Windows Mobile 6 phone that had no internet on it.
Now, I have this high-resolution shiny object that can near instantaneously get any information I want along with _streaming HD video to it_ *anywhere*.
15 years even feels like a stone age. I can't fathom what it has to feel like people in their 60s and 70s.
I'm not quite 60, but it's always interesting to me that I feel quite the opposite of this. When I was 16, I didn't have a computer, didn't have a phone, had never used the Internet, but when I think of how life has changed, it's frankly not much. I woke up this morning, scooped my cats' litter boxes, took out some trash, made myself breakfast, ate that, read some news while eating, then lifted weights in my garage, had some work meetings, wrote up some instructions per a customer request from Friday, and am about to go drive to the lake to go do a 9 mile longboard loop.
That's very close to a normal day in 1996. The biggest difference is I read the news on my phone instead of a physical newspaper. The news was not any more interesting or informative because of that. I guess I can also still do the loop reasonably well, but I'm a lot slower than I was in 1996 when I was a cross-country state champion.
My parents are closing in on 70 and I guess I can't speak for them, but I'm at least aware of the daily routines of their lives, too. Walk the dog, do housework, DIY building projects, visit kids and grankids. Seems much the same, too, with the biggest difference being they're now teaching my sister's sons to play baseball rather than me, but shit, one of her sons even looks like exactly the same way I looked when I was 7! The more things change, the more they stay the same.
If your parents are closing in on 70, I would have expected you to be closer to not quite 50 than not quite 60.
I am just over 50 myself and I agree with your points. Technology has changed but life is largely very similar to wear it was in the 90s. At least day to day. Attitudes are way worse now.
Depends on where you live. My dad is almost 80, grew up in a very rural area, and when he was 16 they'd just gotten indoor plumbing. Up until he was 14, his school was a one-room school house with no heating other than a wood stove. If you were the first kid to arrive for the day, it was your job to get the fire going in winter months.
Housework meant no laundry machine, no dishwasher, and possibly no vacuum cleaner. That means hand washing everything, and beating rugs with sticks and brushes to get the dust off of them.
But the prices haven't been going up by multiples of 6 for the past few years. Things are actually changing now. I don't think it's over, but in the short term, it's going to be considerably more expensive.
Dunno, if in this day and age you are making inference more expensive, more scarce, you are honestly moving in the wrong direction and DeepSeek and others will gladly take your lunch.
If/when it gets to the point where it can replace a skilled worker, the service can be sold for close to the same price as that skilled labour. But the AI can run 24/7, reliably, and scale up/down at a moments notice.
There's not going to be much competition to drive prices down, the barriers to entry are already huge. There'll likely to be one clear winner, becoming a near-monopoly, or maybe we'll get a duopoly at best.
Yes, a lot of people (not me). Why? Well because that was the whole value proposition of these companies, relentlessly pushed by their PR and most of the media- rememmber it was something something Pocket PhDs, massive unemployment etc?
"There's not going to be much competition to drive prices down, the barriers to entry are already huge. There'll likely to be one clear winner, becoming a near-monopoly, or maybe we'll get a duopoly at best."
Based on what exactly? So far every time OpenAI, Anthropic or whatever has released a new top performing model, competitors have caught up quickly. Open source models have greatly improved as well.
I expect AI to be just like cloud computing in general - AWS, Azure, GCP being the main providers, with dozens of smaller competitors offering similar services as well.
I do. "Commoditize your complement". Want to sell lots of silicon? Give away good local models to run on that silicon.
Even if SOTA models in the cloud are a few percentage points better, most work can be routed to local models most of the time. That leaves the cloud providers fighting over the most computationally intensive tasks. In the long term, I think models are going to be local-first.
(Unless providers can figure out a network effect that local models can't replicate).
> In the long term, I think models are going to be local-first.
Why? There's an inherent efficiency advantage to scale, while the only real advantage for local models (privacy/secrecy) hasn't proven convincing for broader IT either.
Just be aware OpenRouter charges a 5.5% fee, I didn’t know until recently. I like the product, and I think the fee is fair, but if you want the absolute best pricing then go direct.
But with open router you can always just use the latest model. If you're committed to eg Claude opus then you're better off going directly to anthropic for sure, but if not, varying other models may be fine too, depending on use case and be massively cheaper. Eg new deep seek model with same mio context window or Kimi k2.6 with 270k context window for subagents which implement
>but if not, varying other models may be fine too, depending on use case and be massively cheaper
Do inference providers have standardized endpoints, or at least endpoints compatible with claude code? Otherwise to pay 5.5% on all your tokens just so it's slightly easier to swap providers (ie. changing a few urls?)
"This change aligns Copilot pricing with actual usage and is an important step toward a sustainable, reliable Copilot business and experience for all users."
I see statements like this as strong indicators that the sales people are wrapping up their work and the accountants are taking over. The land rush is switching to an operational efficiency play.
27x for Opus is genuinely shocking. at that point you're not paying for convenience anymore, you're just paying a GitHub tax. OpenRouter or direct API makes way more sense unless you're really glued to the IDE integration.
Does it effectively bypass regional restrictions for you, so you can use something like the Claude API from unsupported regions such as Hong Kong, or does it still enforce the official providers' geo-restrictions?
OpenRouter is great for budget control, but as they are indirect APIs, your experience with cached tokens may vary, eventually costing much more than in direct depending on the providers.
You can pay with crypto though, which seems to be convenient for people under sanctions or with limited access, or if you are in low-tax jurisdiction (e.g. HK)
That said I think few people using openrouter are actually being selective about providers.
It took half a day to get my opencode setup, was not friendly. A lot of manually cross referencing model and providers. I was actually mainly optimizing for relatively fast providers. It all is super fragile and I'm sure half out of date; I have no idea if these picks are still fast, no promises they are still the same price (pretty terrifying honestly).
I'm mostly on coding plans so it doesn't super affect me. But man is it a bother to maintain.
It's interesting that the cost multiplier for Claude Sonnet 4/4.5/4.6 varies so much (1/6/9), while the API cost is exactly the same for all three models.
Also, the multiplier of 27 for Claude Opus 4.6/4. is way higher than the increase in API price would suggest.
On GitHub copilot you pay per prompt. More powerful models can do a lot more work (consuming a lot more tokens) per prompt. Also, they tend to use more thinking tokens.
One theory of the play of SpaceX might do if everyone migrates to query-based billing:
Provide cheap and unlimited access to Grok for programmers (hence the Cursor partnership/purchase for distribution).
-> This would drag massive revenue right before the IPO announcement, like if the company is super growing
-> At a loss, but don't worry, we need these funds to build the biggest datacenter of the universe.
This announcement would create enough momentum to increase valuation, and because of the merge of his companies, would save his X/Twitter investors from a tragedy.
-> Would also be a great service to Cursor investors and so, who are stuck with their VSCode fork
Which in turn owns Twitter. SpaceX is now a social media company in addition to a rocket company.
One theory I think Matt Levine posited, is that SpaceX will go public with dual-class stock that gives Elon control even with a minority ownership stake, and will subsequently buy Tesla, which doesn't have dual class stock, making SpaceX the singular "Elon Musk company", with him having operational control despite being public.
What's annoying is that it's obvious. In the case of GPT 5.5, if Copilot is going to charge 7.5x what GPT 5.4 costs while OpenAI themselves via the API/Codex only charges 2x of what GPT 5.4 costs, that will immediately raise an eyebrow.
To anybody who's been watching the tech sector with a critical eye for pretty much any period from the late 90s and onward, this is just the enshittification process. For most of OpenAI's existence it's been obvious, to me, that investors were burning insane levels of capital to build the market, and now that folks are locked in, you're seeing higher fees, ads, etc. Yet again, the user is the product; the investors want to siphon your data, attention and once you're hooked, money. And for companies like Microsoft and Apple, those hooks can dig deep.
If you paid attention to the power requirements and amount of hardware being put into data centers, you should have realized that it cost them an order of magnitude more than you were being charged. To rework your analogy: they hooked you, now they're gonna see if they can reel you in.
They can only reel you in if its worth it. I still can code.
And while i do not spend 200$ privat, in my startup we discussed this and our current mental model is, that instead of hiring someone new, we prefer to have more money for tokens.
This is easier for us and has a bigger benefit. The cost of a new / first employee is very high, a 200$ subscription is not. Upgrading that to lets say 400 or 800$ is still alot easier and if i can run multiply and better agents with that money, lets goooo.
I'm looking at education -- teachers and students, not terribly tech savvy, are being mandated to use these tools. And then comes the rug-pull. It was worth it, but now it's outside of their budget. Poorer schools / students can't stay at the cutting edge; richer schools / students can.
“Enshitification” is just when unsustainable subsidies end?
Another reason to hate that word.
From a different perspective, you were granted an incredible gift from the companies who let you use their product on their dime. Hopefully you made the most of it when you had the opportunity.
No, it's much more than that. It starts with unsustainable subsidies, as Uber undermined the taxi industry with a ludicrous burn rate. And then, once everybody's hooked to the point that they can't imagine life without the product, you raise costs. And you iterate: raising costs, lowering quality, selling data, increasing addictiveness. Until everybody wants to get rid of it, hates every aspect of it, but is still hooked to the core product. I'm personally not using these tools, not using uber or Meta products. But I'm still using some Google products and it's hard to extricate them from my life now that I'm using them.
OpenRouter doesn't even have hardware. What are they possibly subsidizing? The platform costs?
OpenRouter is guaranteed to be about the highest margin operator in the business right now. Everyone wishes they'd be them, skimming 5% off as the middleman without any OpEx.
Streaming, caching, and tool calling can get pretty expensive with scale, even when you don't touch inference. Maybe they're doing something clever and are quite profitable.. or maybe they've already taken $40mm from VCs and are currently trying to raise $120mm at a 1.3B evaluation.
They also show headline prices for the cheapest provider of whatever model, but then need to hit different backends some of which may be more expensive. For now they absorb those costs, but the VCs always come knocking.
Just my opinion though. Totally agreed that they have one of the best positions amongst all AI providers from a financial standpoint.
> They also show headline prices for the cheapest provider of whatever model, but then need to hit different backends some of which may be more expensive. For now they absorb those costs, [..]
They do?? I was under the impression I was just playing the price for whatever provider they deemed 'best' for each completion.
Checking now: The way they describe it in their FAQ is that if the price changes, then they will bill you the new price. But I read that as regarding if the primary model provider changes their headline token cost; not in the case of pricing differences for models that have many different backends that host them.
Regardless, I would be more concerned about the streaming costs if the service continues to blow up and they scale aggressively through VC investments. If their 5.5% skim accounted for what they needed, you'd think they could effectively grow organically..
That's so unfair to us hard working developers. A month ago i could buy for .4$ a turn with Sonnet. Now i have to pay at least .9$ for this turn. Weeks ago i could buy for .12$ an Opus turn after they already raised prices and now they want .27$ from me for the same product! They are stealing from us!
They aren't stealing from us, for several reasons. First of all, it's a voluntary transaction. If you don't like the prices, use something else. Or don't use AI at all.
Second, you have no idea what their costs are. It is most likely that they are simply passing on their costs to you. If that was not the setup, users would just go to another service provider who was providing tokens at a cheaper rate. It's not like there is a dearth of competitors in this business.
"Your plan pricing is unchanged: Copilot Pro remains $10/month and Pro+ remains $39/month, and each includes $10 and $39 in monthly AI Credits, respectively."
If there's no discount on credits (in terms of tokens per dollar) over other providers, I'm going to switch to a PAYG provider. If there's a month where there's little to no coding I can pocket the 10$. What incentive do they give to stay with this plan?
Or if you're a business with multiple seats, these plans may be more inefficient than raw API usage billing. Since if anyone at your organization fails to utilize their full $19/39 allotment each month, that's wasting money, whereas with API credits it is 100% utilized.
I don't think they've thought through the implications of this. Everyone should cancel and go usage-based billing with caps.
They do address this in the doc, Orgs can now (although it was vague as to whether it was an option or just the new standard, probably option due to business contracts) 'pool' the Usage billing across all users.
I'm guessing they did that (and the 'temporary bonus credits') to make the pill easier to swallow for that side of customers.
It forces you to pay at least $20 in tokens per user even for people who use less (they probably have stats on how many people use just autocomplete, which doesn’t count against the quota. or have a seat and don’t use the service at all).
This was my first thought too. "Oh cool, I should be seeing lower prices" as I don't use Co-pilot that often anymore. But no, that's not the case. It rather served to remind me that I should probably just cancel.
Everybody who says it's a 5-9-27x seems to not be aware of the obvious loophole. More like 50x increase. You were able to use over $500 worth of Opus on a $10/mo Github plan easily, no hacks. You could just prompt "plan this out for me, don't stop until fully planned, don't ask any questions", and you would get ~$5 worth of planning in one 3x request. At 100 requests/mo, each easily reaching $5, that's easy $500 worth of tokens.
This was my solution to very very though compiler tests that would take sometime up to 4 hours to figure out. Some of the time would be spent on running the tests, but still... I was burning so much tokens. I have free Copilot for my open source work so I wasn't even paying the $20.
Yeah it was crazy. Nowadays I use pi with OpenAI GPT 5.4/5.5, which to me seems both better and more generous than Claude. I supplement it with OpenCode Zen to get access to a bunch of models at token cost, and OpenCode Go ($10/mo) to get subscription-style access to Kimi, GLM and friends.
They're not the only ones in the AI sphere to wind back, but they're the weirdest case in my eyes. Microsoft invests in having engineers building open models and they don't use a single one. I really don't get it.
But what really surprised me most about Copilot is that it would bill you per question, nothing about tokens. So if I managed to produce a prompt that gave me back an insane amount of tokens for something, which using any Claude model would easily accomplish, you were giving me my money's worth, at your own expense. The math is not gonna math out forever.
One of the largest employers publicly engaging in a project which has the outcome of depressing wages. It's easier to "get" if you don't take the trillion dollar gorilla at face value.
This may be a more accurate analogy... "The Porsche you rented at $200/mo now only allows you a maximum of 100km of travel. You will be automatically charged extra when you go over that."
On top of being worth less, the subscriber discounts are gone.
The old plans were $0.033/request for Pro, $0.026/request for Pro+ and $0.04/request for pay-as-you-go. That discount is now gone. They even still advertise "5x the number of requests" for Pro+ over Pro.
Yeah, if I go to a petrol station with 50€, but only get a tenth of the amount of petrol I got last week, I may think that the price has in fact changed.
It’s technically true that the plan prices haven’t changed, it’s just the value you get from those plans has plummeted. It’s classic deceptive sales language.
I was curious why a company would still use the VS Code + Copilot sidebar method for coding, rather than something like Claude Code. Turns out there’s a GitHub Copilot CLI!
I thought I was pretty familiar with available options, but no one in my circles ever mentions this product. It doesn’t seem to have much mindshare.
I'm curious about the opposite: Why would anyone use the CLI when, at least with Copilot, the VSCode plugin is super tightly integrated with VSCode, meaning the agent can see everything I can see. There's no mismatch in linter calls where I can see a lint in the ide that the agent can't find for example. I've had this problem even using CC in their VSCode extension, so I can't imagine it's not an issue in the CLI as well.
Yeah, I've been using it heavily at work since the beginning of January (and have a personal Anthropic sub to compare to). Copilot CLI is pretty good, honestly. Most new features in Claude Code get cloned by Copilot CLI within a couple weeks. Claude models seem mildly more clumsy in that harness than the one they're trained on - subjective guess around 20% more turns for an equivalent task - but it's not a noticeable difference in the final output.
The vs code integration is pretty slick. I can copy and paste function names into the prompt and it automatically turns them into these `#sym:` reference objects that I presume populate the context window with metadata about the function and where it lives. It knows what file I'm currently looking at as I jump around in the code, and that automatically gets loaded into the context. I can also drag and drop folders or specific files for context into the sidebar.
It's a lot of stuff that makes me have to type less into the prompt, since it's already getting so much info from my editor
> I was curious why a company would still use the VS Code + Copilot sidebar method for coding, rather than something like Claude Code.
I use Claude Code, but I kept my Copilot subscription around mostly for really cheap usage of other models when I need to try a different one (which appears to be ending, in a sense) and also the autocomplete in Visual Studio Code which was really great across a bunch of files, I could make changes in one file and then just tab through some others.
I wonder what other good autocomplete is out there.
> also the autocomplete in Visual Studio Code which was really great across a bunch of files <...> I wonder what other good autocomplete is out there.
I am in the same boat. I tried looking for tab/auto-complete implementations ~ a year ago and it was pretty disappointing. If that has changed, would love to know!
I've used it quite a bit. There are a lot of AI terminal coding products and this is another one. It works well, handles sub-agents without issue and does a reasonable job operating in the Copilot ecosystem. It handles mid-task questions and such we well.
But its a really good UI for agentic coding. Not sure why more people don't use it. I've tried the others and keep coming back to Copilot chat. It's a really good tool. Which is why the rugpull on pricing is so concerning.
I’m actually trying to move back from the Claude Code style, I feel like it’s easy to become distant from your own code, and I am feeling uncomfortable with that.
I’ve “vibe-coded” some projects and when I start to find issues or go to refactor them I don’t have that memory of why decisions were made, because many decisions were never made.
The other cool thing is Copilot SDK, so you can build agentic capabilities into apps, or build tools, that leverage the agent harness of the Copilot CLI:
I'm just so confused why people aren't just using ghostty/kitty/terminal.app and claude code. Compared to the other approaches I've tried, it's by far the most effective way to get performance from opus 4.6/4.7
I don't know about others, but I use Copilot more often than other apps because of its tight integration with the VS Code itself where I still spend most of my time working on other things while letting AI do some task that I decided to delegate to it.
That's espoused as the big reason for the price increase: most Copilot subscribed developers it seems have moved to "agentic usage" with the CLI and Cloud-based agents.
Which feels a bit like a kick in the pants for me as a developer that was primarily using Copilot for VS Code ghost text and very rarely used the Chat sidebar much less "agentic" tools.
Copilot Pro sort of made sense for my personal account when amortized across a year, but I don't want to "waste" $10/month on credits I won't use most months.
I've been using it for a few months because Copilot was the only AI blessed by our corporate overlords. It's not bad, I would say it's about 80% as capable as Claude Code, which I've used extensively on personal projects. However CC was recently approved, and I'm betting that with these changes to Copilot pricing we'll end up dropping it like a hot potato.
I tried the VS Code + Copilot sidebar approach a few months ago. It was definitely rough around the edges compared to Cursor/Claude. In our corporate environment, we weren't even able to use frontier models.
Search has become so bad that I also struggled to find Claude Code alternative and made my own tight (not editors, not plugins, not agents, strictly similar to Claude Code CLI) list: https://github.com/omarabid/cli-llm-coding
The list is not long but there are quite a few options. Even Grok has its own CLI!
The reality is, even though a CLI prompt looks very simple, it's a very complex piece of software. I personally use Claude Code (with GLM) and anything else I have tried was significantly inferior (with the exception of opencode).
> In March 2026, Windsurf replaced the credit-based system with a quota-based usage system. Instead of buying and spending credits, your plan now includes a daily and weekly usage allowance that refreshes automatically.
With hindsight, per-request pricing makes no sense at all if an agent can burn a widely varying amount of tokens satisfying that request. These pricing plans were designed before coding agents changed the dynamics of token usage.
I wouldn't call it hindsight - I don't think anyone, at any stage, thought running a 10 minute+ sonnet session for 1 premium credit was ever profitable. We all knew it was a loss leader to get people using it.
It would have been profitable if that premium credit cost more than a negotiated discounted rate with Anthropic. We have no way of knowing if there were negotiated rates though!
There is no way to make that cost model profitable consistently. If 1 prompt can mean 100's/1000's of requests over hours, and you only pay for that 1 premium prompt, that can never be profitable.
Guys, you're discussing a house of cards to begin with: No matter how you're paying for the $CURRENTSOTA you're not garunteed that next month what you pay for will be the same.
So, lets do some honest evaluations:
1. The model itself is a non-deterministic engine of work with an unknown value; it's real value is just magic.
2. The business model itself is non-deterministic engine of profit with a known value; whatever the VCs have put into it, _must_ be piulled out. If Ed Zitron's numbers are correct, circa 2030, it's several trillion dollars.
So do some matrix multiplication of non-determinism vs determinism, and realize that the value proposition for _you_ is only going to decrease because #1 can never outpace #2, ensuring enshittification captures a smaller and smaller whale.
We know this. This has been the last 2 decades of money extraction from software. It was ok when it was some 12 year old's parents CC. But now it's you, or your business, that's going to either ben squeeze for value or squeeze out of the market.
And everyones squabbling about the color of the cost.
ok
The problem with assuming that tokens can only get more expensive is that the Chinese open weight LLM firms have dropped models which have a known, fixed price that can never get more expensive (since we can run them on hardware we own).
I wonder if GitHub (Microsoft) is implicitly betting that enterprise demand is sticky enough to absorb these rates, especially given that Opus 4.6 “fast” was being listed at a 27x multiplier. Maybe they saw enough usage at that price point to conclude the demand is real. Or maybe the strategy is to keep the enterprise customers who can justify it while shedding heavier individual and power-user usage.
The interesting question is how long it takes enterprises to notice the capability/pricing tradeoff, and whether they respond by limiting access to the strongest models internally.
The part that worries me is that this market is still very early. Most developers and organizations are still learning how to use these tools effectively. Raising the experimentation cost this much may slow down the discovery process that makes the tools valuable in the first place.
I also suspect that there are many "slow-moving", Microsoft heavy enterprises but with in-house devs that can't get anything but Copilot approved, and Microsoft trusts this will remain so.
It's not turning consumption based because there are a ton of these licenses just sitting idle.
As a single data point, this is absolutely true. At my current "Big Corp", Copilot was immediately approved while Claude is entering month 2 or 3 of trying to get approval.
Additionally, we got copilot for every user, including those that never write code or use AI tools.
Subsidies stop when LLMs improvement plateaued (though they still benchmark higher somehow). At some point, you have to make money or at least break even; and I think they concluded that we reached that point.
As someone that is on the enterprise side in a non-tech F500 company, what I'm seeing is some FOMO and need to be part of the hype cycle. We're about to plonk a bunch of money on more Copilot licenses. Something got in the water where all the C-levels the past two months are pushing everyone to use AI but when they bring up examples of their uses its like "I use it to rewrite my emails" or prompt 'engineering' ideas that point more to patching over poor processes, data management, and decision-making within the organization or not.
What we're seeing across the board is every software company tossing AI onto their name or sales pitch and no one understanding what that actually means. But we will spend money on it because of FOMO.
I really question if we're reaching the end of the hype cycle to the point. I wish I were brave enough to put money on it. It feels like there was a command from up top to 'do something with AI' and leadership is scambling for some resume-building projects vs doing the hard work they should've done the past two years at a people and process level.
I really hope that they think so and that they're wrong and they get burned hard. Them and all the AI labs that lied, stole, inflated, hoarded and tried to justify all this as an existential moment where AGI would radically change society. I hope their calculus to reel in paying users is all wrong and now they all crash and burn instead of recouping VC money.
I was always curious how they can sustain a request based pricing model when requests can range from tiny to huge with all the modalities GH Copilot offers. Was a steal for agentic coding that turned out to be too good to be true, in the end. Still: Thank you for the ride.
I don't use Copilot or any paid AI but all of this usage-based billing reminds me of cellphones back when you paid per individual text message.
Usage paying for AI is 1000x crazier because you're not even getting a guarantee in the thing you pay for in the end. You have to keep feeding it prompts and hope it gives you the solution you want. You may end up with no expected result yet you are paying for it. At least with texting, you got what you paid for.
I wonder how long it'll be before all AI costs are flat unlimited monthly fees or even free across the board, without compromise.
It incentivizes you to do most of that prompting on your own hardware/time, and only feed the final prompt with only necessary context to the big AI in the sky. It might even force you to think about the problems yourself for a bit!
I expect in the future we'll find out that someone in the industry was juicing the numbers with fake thinking tokens or something. The whole pricing model of charging you for the tokens it generates while not knowing how much it is going to generate going in has always been pretty crazy.
Yeah, this was my frustration with Suno and Sora. You can burn a lot of credits (not to mention time) generating things that aren't what you wanted.
I don't mind a PAYG model for a simple chat interface. But when it comes to actually producing things, you burn through TONS of tokens creating the wrong output.
> I wonder how long it'll be before all AI costs are flat unlimited monthly fees or even free across the board, without compromise.
That's already the case if you can self-host an LLM; you don't even need a mythical H200: gamer-grade GeForce cards can get you a long way there (if this page is to be believed: https://www.runpod.io/gpu-compare/rtx-5090-vs-h200 )
...after RAM prices return to normalcy, of course - and then wait another 2 or 3 generations of GPU development for a 96GB HBM card to hit the streets - and also assuming SotA or cloud-only LLMs don't experience lifestyle-inflation, but I assume they must, because OpenAI/Anthropic/Etc's business-model depends on people paying them to access them, so it's in their interests to make it as difficult as possible to run them locally.
Github had, by far, the most easily game-able agent usage policy. People would force the agent to run a script before the end of turns that consisted entirely of `input("prompt: ")` so that you could essentially talk endlessly to an agent for the price of a turn. I see this less about the future of this industry and more about fighting the costs incurred by bad actors.
I never played any games like that, but simply giving the agent a clear exit criteria and instructions to check the exit criteria every time it thinks it's done on a complex task was often enough to keep it chugging away for most of a day on a single prompt in my experience. Per-prompt pricing just isn't sustainable period, even if everyone is acting in good faith.
I once asked it to do a comprehensive security review of our code. It churned for nearly an hour (and then produced 90% false positives). Insane that that usage was charged the same amount as me just saying "Hello".
One reason I used it was that I wasn't locked into a single provider and switching them was as easy as changing a drop-down. Small feature? Sonnet or GPT5.4/mini? Large changes? Opus. And why not see how good Raptor Mini does this one refactor?
It also helped build an intuition of what wach model could do and which parts it was weaker at because you could try them almost side by side, especially if one model's output wasn't great.
That said, these were all side projects so nothing truly consequential. Otoh, you might leave some extra perf on the table but I found the models worked quite with the Copilot harness.
Yeah, this is a very useful abstraction layer. The entire concept of separating the model creator from the model runner is good for competition and is customer friendly. Which means they likely hate the concept and want to kill it.
Gosh, imagine getting to do that with your TV/Streaming subscription. Getting to pay one fee to access some set number of hours per month from any of the providers.
The problem is I can't afford the tokens! Even on my $10/mo plan, running either 100 opus, or 300 sonnet agent runs would cost hundreds of dollars - well above my budget!
I started to use github copilot with vscode, but have never been too happy about the system. Over the months I gravitated to much more agentic workstyle, hardly ever editing much code by hand. The vscode IDE was getting more in the way. I had already started to look at OpenCode, and when I found it has a web interface, I was happy to switch over. I use a simple editor (KDE's Kate), or just less to skim through the code and/or a git diff.
OpenCode has some free models in it, but I think I will need to get some kind of subscription for a better one. But it won't be copilot any more. The market is moving so fast that I don't know what are the most resonable models, or the most flexible way to set them up so I can switch when prices change yet again.
So given that I primarily interact with LLM's through VSCode, and I prefer the Copilot interface to the Claude Code plugin, does anyone have any suggestions on other plugins I should try? In my experience, Copilot is much more "plugged in" than any of the other plugins, in the sense that it can see things like linter outputs in VSCode. Basically, copilot "sees what I see" in a way that no other plugin or command line tool can, which make it much more ergonomic to use.
With this pricing change, I see no reason at all to stick with Copilot in principle, but I really need to solve this issue of IDE integration to move on.
You can use Copilot Chat* with basically any API provider, and if you switch to the VS Code Insiders build you can configure it to use literally any OpenAI API-compatible endpoint.
Other than that Zed has a similar experience which is pretty decent.
* By which I mean the good one, whatever it's called now - the part of Copilot that used to be a plugin and is now part of VS Code, not the thing that has always been part of VS Code.
What's the current situation for coding with Local LLM's on decent hardware? I have an M3 Max with 64 gb of ram and am thinking I should start looking at Ollama and Opencode? Is this a useful stack for smaller personal projects?
It’s getting there. You could give a try with qwen 3.6. It’s worth paying for better models in the cloud, but local models are now better than nothing.
One nice development recently was ollama's support for MLX optimization on Mac hardware. It's not obvious how to know you're using a model that works with it, yet, so it's rough around the edges.
On my personal account, Copilot Pro+ still only gave me back Opus 4.7, whereas my work's Pro account still lets me use Opus 4.6.
So, my gut says, it's entirely possible that Pro+ will continue to have more segregation on model availability...
FTA
> Last week, we also rolled out temporary changes to Copilot Individual plans, including Free, Pro, Pro+, and Student, and paused self-serve Copilot Business plan purchases. These were reliability and performance measures as we prepare for the broader transition to usage-based billing. We will loosen usage limits once usage-based billing is in effect.
There's enough weasel wording here that I would expect only certain models get re-enabled on Pro.
e.x. lots of people seem to get good enough results from Opus 4.6, personally I prefer it over 4.7 in GH Copilot... locking that down to Pro+ would be, given this salvo of enshittification, a 'logical' move on their part.
The cheapest copilot plan felt totally unsustainable to me. For around £8 month i was getting 100 opus 4.6 prompts (albeit with a reduced context window size around 128k iirc vs 200k to 1m for first party hosted opus). Gpt5.4 was hosted with 400k context iirc.
On top of that, you’ve got 2000minutes of container runtime, so running cloud agents was included. As was anthropic agent sdk mode via copilot which is very comparable with claude code - not identical, the anthropic “modular prompt” is much leaner in the sdk version.
I cant say im mad, i got above what i paid in value. That said, going forward ill probably go back to openrouter payg rather than a subscription.
I got a free 3months of the gemini £19 plan and ive been playing quite a bit, 3.1 pro is a good model, i just find it slow. Flash i think i under appreciated until now.
I pay for Copilot annually, and mostly for its code auto completion features. I use CC if I want to do anything agentic. Not sure if I want to pay more for occasionally-good-intellisense at this point.
But you can no longer amortize annually, which makes it even more a question of "is this worth it this month?" each month. Especially for personal accounts.
I'm similarly thinking about sticking with the auto-downgrade back to Copilot Free when the annual sub ends and then just yelling about it any months I hit the 2000 completion cap.
I wouldn't mind a plan between Free and Pro that is just "all I care about is code completion and next edit suggestions".
I liked copilot because I didn't have to think about tokens. I get hung up when having to think about the price of things, and its hard to think about the project at the same time I got to think about token usage like a gas bill. The usage system had its own issues, but having a set amount of requests was a very comfortable way to use a paid AI service.
Not paying per token? Not sending my code to someone else's servers for inference? That's the stuff of sweet dreams for a stingy, paranoid solopreneur like me.
If I could run a local model comparable to even Sonnet 4.6 without shelling out $50K in hardware, I'd do it in a heartbeat. But all I have is a 32 GB of RAM and an old RTX 4080.
Or am I not up to speed? Are there decent coding models that can run on dev laptops? Not that that's what you were suggesting by recommending a local model, necessarily; just curious.
I do love using local models when I can, but qwen-35B is the best model I can run, and while its an insanely good local model, it does not compare to the big ones.
Looks like Microsoft has run out of compute and can't scale it fast enough to serve copilot users and Azure AI Foundry needs, given that the customer base is growing there as well.
I think that only applies to held-over users on the annual plan:
> Users on annual Pro or Pro+ plans will remain on their existing plan with premium request-based pricing until their plan expires, however, model multipliers will increase on June 1 (see table).
It isn't just the big multiplier increase, they also say "...and no new models or features will be added to annual plans going forward."
Can you imagine ten months from now and you're still rolling Sonnet 4.6?
Cancel/refund is looking pretty good. They're doing refunds until May 20.
"To request a refund, go to Settings → Billing and licensing → Licensing, select Manage subscription, then choose Cancel and refund "subscription". (The phrasing varies slightly depending on your subscription ). This option will be available until May 20."
It is an apples for apples comparison since those new multipliers only count if you are on an annual plan in which case the premium request system stays in place until you either cancel and get a refund or until your renewal comes up. https://docs.github.com/en/copilot/concepts/billing/usage-ba...
Those multipliers will only apply if you are currently on an annual subscription (and only until your renewal comes up or you cancel). So I assume they simply want to make it as unattractive as possible to get most people to cancel it and move to the token based system.
That's not an answer. It's specifically a discrepancy between 5.4 and 5.4-mini. If you look at all other models/generations you see that the cheaper model indeed has a lower multiplier. It's very strange that only 5.4 doesn't have this.
In places with reasonable consumer protections (Australia, Germany) it almost certainly is illegal unless they give a full (whole year) refund. I think the short time limit of applying for a refund won't be looked at favorably either. Regardless of their ToS which I'm sure covers this.
But companies do lots of illegal things, and in general nobody takes them to court over it.
I doubt you can force them to provide the service with the original terms, but you might be able to ask for a (partial) refund. If not today, after a week of verbal abuse they will receive for this online.
It depends where you’re located. In the EU they have to honor the contract you entered, but presumably there is a clause that they can prematurely terminate the contract without cause and give you all of your money back (from the start of the contract).
That kind of clause would be void in many places around the world.
For example, the German Civil Code states:
Section 308 - Prohibited clauses with the possibility of valuation
In standard business terms, the following in particular are ineffective:
[...]
4. (Reservation of the right to modify) the agreement of a right of the user [TL note: this means beneficiary of the terms, eg. party or other subject of the contract] to modify the performance promised or deviate from it, unless the agreement of the modification or deviation reasonably can be expected of the other party to the contract when the interests of the user are taken into account;
My thought exactly! First the usage limits + model limitations and now fundamental change to the billing. Hope some consumer watchdogs are looking into this!
I guess this is the "Google App engine" of Vibe coding when google raised pricing of App Engine significantly after being in preview. Doing weird price changes like this when coming out for new services make much more sense than doing this change which will just anger users.
What's the residual value of Copilot after the changes? For Enterprise plans, even copilot code reviews will be charged at token price + github action minutes for the execution of the review. One can roll one's own reviewer and have it spend the API tokens as well. At least one will be able to select the subset of files for the change set, if needed.
The background agents will also depreciate in value because of their harness that's a black box that's not optimized for token usage at all. Rolling one's own will be a better choice here.
so what's everybody using to get autocompletions in vscode? i've been using copilot just because $10 is cheap, but i use opencode for everything other than completions.
i tried the continue vscode extension, and it seemed kind of janky. are there better options?
That was the first thing I turned off in VSCode. Autocomplete for my TypeScript projects was great. And the "AI" suggestions/completions were really getting in the way of me still being the "driver."
I'd be fine with it if they can make Sonnet 4.5 unlimited. I haven't personally seen any major differences between 4.6 or any of the other newer models. Claude 4.5 seems to be the right balance and works great for me.
"Paid for annual? Tough luck, from now on your usage limits are reduced by 89%. You can do 11% of what you paid for. Good luck if you paid annual a month ago!"
And then they have the gall to say
> "The bottom line: Plan prices aren’t changing"
If anyone lives in a place like Germany or Australia and has an annual sub, please take them to court, you're guaranteed to win because you have reasonable consumer protections and their ToS doesn't stand a chance. 9x reduction is unreasonable and the consumer cannot be expected to see this coming.
In case some diehard enshittifier believes that consumers should know better and businesses should be allowed to get away with it, where is the line? 99% reduction? Is that still okay?
If this situation is to be acceptable then it should be regulated as a financial product like stocks, which come with knowledge tests of "do you know you can lose all of your money?". And come with regulatory compliance and all that.
According to `bunx ccusage` I'm easily doing $250-400/day in "real" API costs on my $200/month plan. There's no way everybody else isn't going to do the same thing and completely change the industry again. Both beginner and advanced developers are already hooked on all this stuff and they all know it.
"It" being the end of subsidization of tokens and plans (expected) but while lock-in to foundational models and cloud services is still lacking. Guess investors want their ROI sooner than later, given how big of a wrench the AI boom has thrown into global economics.
Built credit pricing into my SaaS for AI features and the hardest part wasn't the math, it was that customers can't easily predict their own usage. They underuse and feel cheated, or overuse and churn. Subscriptions hide that volatility from the customer. Usage based pricing makes it their problem, which is honest but harder to sell.
We all knew this from the very beginning but couldn't compete with OpenAI or Anthropic on their subscription-based pricing strategy. It was nuts except for those few corporations burning investor money to keep competition out as long as possible. Now they don't have to hide anymore that subscription pricing won't do it for ai. The pyramid scheme is falling.
The problem is that people expect to get the output of 100 people with a $20 subscription by spawning multiple agents. This is unrealistic. I'm using 2 codex plus account and able to manage a repo with 265-300k lines of code.
The point is - if its the same or more expensive per month than a real human employee - why pay for AI ?
Human retain knowledge, product knowledge, can pick up more work often for the same money. And having many of them means your business wont go down if provider suddenly bumps API pricing.
How much is the average pay for a junior developer in US? Its definitely much costlier considering PF and other benefits. Just the math. if you use it efficiently it's much cheaper than hiring a permanent staff. You can maintain a lean team and do all the mundane boilerplate coding with AI.
some of Github's open source maintainers have lost their free github copilot pro, guess this is really the next step for them to save cost in their infrastructure.
Whose idea was this “premium request” model anyway? If you’re going to invent a new metric used to bill, why not align it with what, even at the time, was a clear underlying cost structure that GitHub actively chose to ignore for a more confusing system.
It made more sense in the ye old days where a request was basically just a chat message in a sidebar and it could also edit code. Then saying someone can use 300 chat messages a month kinda makes sense.
Turns out when a request can spawn tens of subagents and use millions of tokens over many turns of toolcalls then suddenly github copilot has a massive financial problem on their hands.
This approach started with the “Ask a question about your code” feature, which is more comparable to single chat message with relatively predictable token usage. Now it’s an agent who might work for 30 minutes, read the whole codebase, and write 1000 lines
I'm not usually a Conspiracy Guy, and the answer is probably `incompetence * tech_debt`. But I think that having sufficient layers of abstraction to any billing model is a useful way to hide the real cost of things. It's why it's done everywhere.
As a Github Copilot user, who mostly just uses chat in the VS Code editor but still burns through my Pro limit every month -- what's the best alternative price to performance? Claude Code?
It may already be cheaper if you calculate the hidden cost accumulated by LLM generated tech debt. At least some of it could be avoided by manually doing it better to begin with.
cursor, windsurf, and CC are all already on usage-based models so I guess what really matters is whether Copilot's GitHub integration depth justifies the price per token vs the alternatives
I'm happy I invested in local solutions and cutting context to the bone for API providers. Claims about AI being able to fully replace programmers never took into account the long-run equilibrium price of inference.
That's my point. They made those decisions without any consideration for the long run, which would have required them to project the cost of AI services years into the future. Obviously management didn't do that and had no way to do that. It made current earnings look good, though, which was enough for them when they made the decision.
I'm not sure I understand this. All I know is now, I pay $39/month (actually less because I paid a year up front), use the agent, mostly on auto--and only choosing a model if it got stuck or in a loop--every day, and haven't hit any limits yet. It seemed to good to be true, after hearing others talk of $300/month bills. I guess it was.
This subsidized inference is just a marketing ploy to increase prices and profit.
If common people can have a DIY setup with an open source model cheaper than those behemoths with a scale advantage, it's clear that we have been played.
Time to either self host a Chinese open source model or to just pay the cheap Chinese providers.
Yeah, local is clearly the future. Even beyond the cheap Chinese models you can install the apfel[1] stuff if you're on a mac and want a quick available onboard cli option. And I'm sure people will adapt the Flash-MoE[2] integration to be even better soon as well.
DeepSeek (and other open weight models) can be served by anyone with the means to do it. All of the big open weight models are available on OpenRouter via commodity providers.
Glad this was announced because I didn't even realize that our (small) team has been paying $20/month for Github Copilot when none of us are using it, so used the opportunity to cancel CoPilot altogether. I think it was free when I first activated it (this was also before Claude, Codex, Gemini, and while not that great, it was why not), and didn't realize it had switched to paid and bundled with our Github bill, and now per usage.
Re 1: Current models don't solve coding. They are useful tool for it though.
Re 2: Open weight models seem to be less than a year behind proprietary ones, so sure, if you're willing to spend tens or hundreds of thousands of dollars on a super computer that you probably don't fully utilize instead of renting time on someone else's super computer for a lot less.
People need to wake up and stop being surprised by these billing increases. I see it on every update of every model. This was all subsidized by VC and company money. Now they need a return and the prices will keep going up. Be glad that you took advantage of that up until now, but can we stop the pearl clutching when we all know the amount of money being dumped into AI and the lackluster returns?
It's less surprise, but more confusing given the game theory as their competitors are not doing the same thing and the multiplier changes alone will likely churn current users.
tldr: people were running multi-hour agentic coding sessions for the same flat fee as a one-liner autocomplete, github was eating the bill, and that party's over on june 1st
Something is hilariously off here: Why should I pay $10 and be forced to use it by the end of the month, while I can pay $10 and have it last as long as I want?
Their "API pricing" is exactly the same as that of providers: https://docs.github.com/en/copilot/reference/copilot-billing...
I'm thinking the same. Downgrade to Pro and use OpenRouter (same price) for overage.
Seems a massive loss for Microsoft. Presumably there's a further rugpull to come.
> Presumably there's a further rugpull to come.
How would that be? They are already charging as much as the underlying providers. They can hardly expect to have any customers if they are charging more.
Enterprise sales will be the answer. Microsoft will have some story that convinces an exec eight levels up the org chart from the normal users that this is an essential product they need to overpay for. Given their existing relationshipsand immense sales team they'll probably have success.
That story is data governance. Corporate already have a data-agreement with MS, storing all their data there. Github copilot is covered by that, while a individual agreement with e.g. anthropic needs lawers involved.
It’s precisely this, and, to be fair, it’s a rational approach given a Data Security Exhibit starts at 6 weeks and can hit 6 months to complete. That being said, I work with regulated data, so YMMV.
Microsoft is simply the default answer for most large corporations. Getting access to some Microsoft subscription is very easy, because of the existing framework agreements, Microsoft providing any and all compliance slopuments needed and already being pre-cleared for corporate data etc. Meanwhile trying to use another provider (e.g. Anthropic) would be a one year endeavor, minimum.
Can confirm, this is the situation.
I'm already on Pro. Why should I keep it?
e.g. if on an annual plan? 0x will be gone, but there are okay 1x and 0.3x models left. I am pretty much curious how the early may test invoicing will look like. current setup of tools etc. is way too chatty eats up 1+M token per PRU easily. not sure how much is cached.
private repos
Can you elaborate please? I still clone my private repos and work on them using OpenRouter Credits and OpenCode (or Copilot itself: It supports OpenRouter BYOK). No?
Copilot pro, not github pro
I have to wonder if it's because of how many Enterprise customers they have who have standardized on Github Copilot and gotten it through the gauntlet of legal approvals etc.
This rings true to me as someone who's worked at a few large corps like this. A price hike does not change things when there is a mandate to use MS products over other vendors.
I'm wondering if they're basically saying they're going to give $10/month free API credits to students and open source maintainers and so on... while otherwise getting out of the consumer portion of this space.
they're downsizing free github copilot pro for open source maintainers. At the very least, it looks like small open source projects got their free copilot pro cut off
Enterprise gets pooled credits and will like having everything go through one place so I think it still works.
You can pool credits through open router (afaik, I'm only using a single user account), but if you top-up $10 per user, per month, any unused credits will rollover.
Tbh I think it still works, but only because the new allowance will likely get used very quickly within a billing cycle - I'm expecting this change to increase our orgs bill significantly based on how many API credits with open router I consume in a weekend using a single agent in a pairing style.
The pooling will only be useful if you have a bunch of infrequent/low usage users that you still want to have licenses.
Is there a way to use the autocomplete feature with an api?
The era of subsidised inference is truly ending. The new model multipliers (https://docs.github.com/en/copilot/reference/copilot-billing...) seem like a huge leap, though. From 1x to 6x for new-ish GPT and Sonnet models. 27x for Opus...
Seems like folks would be better off with OpenRouter instead.
Lots of us have noticed that usage limits for Claude have been nerfed in recent weeks/months.
If anything, these new multipliers are more transparent than anything OpenAI or Anthropic have communicated regarding actual costs and give us a more realistic understanding of what it's costing these providers.
The fact that we were able to get such a substantial amount of usage for $20/$100/$200 a month was never meant to last and to think otherwise was perhaps a bit naive.
This feels like a strategy from the ZIRP era of tech growth where companies burned investor capital and gave away their products and services for free (or subsidized them heavily) in order to prioritize user acquisition initially. Then once they'd gained enough traction and stickiness they'd then implement a monetization strategy to capitalize on said user base.
However, inference costs for entirely good enough models are likely to keep declining in the future. We're probably hitting diminishing returns on model size and training. The new generations aren't quantum leaps anymore, and newer generations of open source models like DeepSeek are likely to start getting good enough.
There's going to be a limit to how much they can raise prices, because someone can always build out a datacenter and fill it up with open source DeepSeek inference and undercut your prices by 10x while still making a very good ROI--and that's a business model right there. Right now I'm sure there's a lot of people who will protest that they couldn't do their jobs with lesser models, but as time goes on that will get less and less. Already right now the consumers who are using AI for writing presentations, cooking recipe generation and ELI5 answers for common things, aren't going to be missing much from a lesser model. That'll actually only start to get cheaper over time.
Also for business needs, as AI inference costs escalate there comes a point where businesses rediscover human intelligence again, and start hiring/training people to do more work to use lesser models--if that is more productive in the end than shelling out large amounts of cash for inference on the latest models. [Although given how much companies waste on AWS, there's a lot of tolerance for overspending in corporations...]
> because someone can always build out a datacenter and fill it up with open source DeepSeek inference and undercut your prices by 10x while still making a very good ROI-
Not sure how it all works out. Currently trillion dollar companies can't make a native app for platforms. Everything is just JS/Electron because economics does not work for them.
And here companies can make GW data center running very expensive GPUs for 1/10th of current prices. Sound little fanciful to me.
The price you pay for anthropic must include the price of training new and better models which is incredibly costly. If you use the models someone else already spend money to develop you don’t need to pay this price.
I think so too.
And at some point even frontier model costs will hopefully come down (if there is still a meaningful difference between closed and open source models at that point) as all of the compute that's being built out right now comes online.
I hope it's true, but right now hardware prices are insane
I guess the new models will still be quantum leaps, but literally: "The smallest possible change in a system"
Yups... Mythos is the smallest possible leap. Not a standard model generation advance, not even a version point advance. Just the smallest possible quanta of a change. We are absolutely hitting a plateau any day now. Any day. Any time. Any second now. Yup. Right now! Surely!
Yeah. AI progress is insanely fast if you compare it to anything else. Where else is a one year old technology already hopelessly outdated? 10 years ago is basically stone age.
I am continually tripped out by the fact when I was 16, I didn't have a 'smartphone' beyond a Windows Mobile 6 phone that had no internet on it.
Now, I have this high-resolution shiny object that can near instantaneously get any information I want along with _streaming HD video to it_ *anywhere*.
15 years even feels like a stone age. I can't fathom what it has to feel like people in their 60s and 70s.
I'm not quite 60, but it's always interesting to me that I feel quite the opposite of this. When I was 16, I didn't have a computer, didn't have a phone, had never used the Internet, but when I think of how life has changed, it's frankly not much. I woke up this morning, scooped my cats' litter boxes, took out some trash, made myself breakfast, ate that, read some news while eating, then lifted weights in my garage, had some work meetings, wrote up some instructions per a customer request from Friday, and am about to go drive to the lake to go do a 9 mile longboard loop.
That's very close to a normal day in 1996. The biggest difference is I read the news on my phone instead of a physical newspaper. The news was not any more interesting or informative because of that. I guess I can also still do the loop reasonably well, but I'm a lot slower than I was in 1996 when I was a cross-country state champion.
My parents are closing in on 70 and I guess I can't speak for them, but I'm at least aware of the daily routines of their lives, too. Walk the dog, do housework, DIY building projects, visit kids and grankids. Seems much the same, too, with the biggest difference being they're now teaching my sister's sons to play baseball rather than me, but shit, one of her sons even looks like exactly the same way I looked when I was 7! The more things change, the more they stay the same.
If your parents are closing in on 70, I would have expected you to be closer to not quite 50 than not quite 60.
I am just over 50 myself and I agree with your points. Technology has changed but life is largely very similar to wear it was in the 90s. At least day to day. Attitudes are way worse now.
Depends on where you live. My dad is almost 80, grew up in a very rural area, and when he was 16 they'd just gotten indoor plumbing. Up until he was 14, his school was a one-room school house with no heating other than a wood stove. If you were the first kid to arrive for the day, it was your job to get the fire going in winter months.
Housework meant no laundry machine, no dishwasher, and possibly no vacuum cleaner. That means hand washing everything, and beating rugs with sticks and brushes to get the dust off of them.
It does feel like the music is about to stop.
It has been years now, of cash injections, investors can't keep feeding the beast forever.
It has been years now of reading this same comment... Surely people can't keep typing it forever.
But the prices haven't been going up by multiples of 6 for the past few years. Things are actually changing now. I don't think it's over, but in the short term, it's going to be considerably more expensive.
Dunno, if in this day and age you are making inference more expensive, more scarce, you are honestly moving in the wrong direction and DeepSeek and others will gladly take your lunch.
Did anyone really expect AI to be cheap?
If/when it gets to the point where it can replace a skilled worker, the service can be sold for close to the same price as that skilled labour. But the AI can run 24/7, reliably, and scale up/down at a moments notice.
There's not going to be much competition to drive prices down, the barriers to entry are already huge. There'll likely to be one clear winner, becoming a near-monopoly, or maybe we'll get a duopoly at best.
> Did anyone really expect AI to be cheap?
Yes, a lot of people (not me). Why? Well because that was the whole value proposition of these companies, relentlessly pushed by their PR and most of the media- rememmber it was something something Pocket PhDs, massive unemployment etc?
"There's not going to be much competition to drive prices down, the barriers to entry are already huge. There'll likely to be one clear winner, becoming a near-monopoly, or maybe we'll get a duopoly at best."
Based on what exactly? So far every time OpenAI, Anthropic or whatever has released a new top performing model, competitors have caught up quickly. Open source models have greatly improved as well.
I expect AI to be just like cloud computing in general - AWS, Azure, GCP being the main providers, with dozens of smaller competitors offering similar services as well.
I do. "Commoditize your complement". Want to sell lots of silicon? Give away good local models to run on that silicon.
Even if SOTA models in the cloud are a few percentage points better, most work can be routed to local models most of the time. That leaves the cloud providers fighting over the most computationally intensive tasks. In the long term, I think models are going to be local-first.
(Unless providers can figure out a network effect that local models can't replicate).
> In the long term, I think models are going to be local-first.
Why? There's an inherent efficiency advantage to scale, while the only real advantage for local models (privacy/secrecy) hasn't proven convincing for broader IT either.
Yeah, totally. The recent pricing changes have just made my Copilot subscription go from great deal to awful value over night.
I've been wanting to get off MS more generally and this is good motivation. Will be playing round with OR this week.
Just be aware OpenRouter charges a 5.5% fee, I didn’t know until recently. I like the product, and I think the fee is fair, but if you want the absolute best pricing then go direct.
But with open router you can always just use the latest model. If you're committed to eg Claude opus then you're better off going directly to anthropic for sure, but if not, varying other models may be fine too, depending on use case and be massively cheaper. Eg new deep seek model with same mio context window or Kimi k2.6 with 270k context window for subagents which implement
>but if not, varying other models may be fine too, depending on use case and be massively cheaper
Do inference providers have standardized endpoints, or at least endpoints compatible with claude code? Otherwise to pay 5.5% on all your tokens just so it's slightly easier to swap providers (ie. changing a few urls?)
> Do inference providers have standardized endpoints, or at least endpoints compatible with claude code?
Yep, you can plug deepseek/kimi/minimax into claude code just fine. Or run everything through another harness like opencode instead.
Wow thats a lot for routing traffic.
And handling API tokens, and billing, and reliability, and middleware. I am not affiliated with them but it’s not “just” routing.
Apple still charges 30%. 5.5 seems pretty reasonable. /shrug I dunno.
Payment processing likely eats up at least 2-3% of that
IIRC OpenRouter charges you for the payment processing fee also.
Still worth it IMO to be able to switch from Provider A to Provider B if Provider A is having a bad day.
"This change aligns Copilot pricing with actual usage and is an important step toward a sustainable, reliable Copilot business and experience for all users."
I see statements like this as strong indicators that the sales people are wrapping up their work and the accountants are taking over. The land rush is switching to an operational efficiency play.
The sooner the better. Let's take a look at the long term, enshittified, viable product before we get too dependent on the trial version.
And enshitification starts.
Even Sonnet 4.6 is 9x multiplier (previously 1x)!
The only model I even used on Copilot was Sonnet and now its got a ridiculous multiplier.
At this point they might as well just charge per Million tokens like every other provider instead of having a subscription.
> At this point they might as well just charge per Million tokens like every other provider instead of having a subscription.
Pretty sure that's what they will eventually do
... that is exactly what they will do. Just click the link in this thread, or read the headline.
Why the multipliers then at all?
The multipliers are there only for current annual plan customers. After 2026 its all tokens.
27x for Opus is genuinely shocking. at that point you're not paying for convenience anymore, you're just paying a GitHub tax. OpenRouter or direct API makes way more sense unless you're really glued to the IDE integration.
I keep seeing people mention OpenRouter.
Does it effectively bypass regional restrictions for you, so you can use something like the Claude API from unsupported regions such as Hong Kong, or does it still enforce the official providers' geo-restrictions?
OpenRouter is great for budget control, but as they are indirect APIs, your experience with cached tokens may vary, eventually costing much more than in direct depending on the providers.
You can pay with crypto though, which seems to be convenient for people under sanctions or with limited access, or if you are in low-tax jurisdiction (e.g. HK)
Caching is advertised per model+provider.
That said I think few people using openrouter are actually being selective about providers.
It took half a day to get my opencode setup, was not friendly. A lot of manually cross referencing model and providers. I was actually mainly optimizing for relatively fast providers. It all is super fragile and I'm sure half out of date; I have no idea if these picks are still fast, no promises they are still the same price (pretty terrifying honestly).
I'm mostly on coding plans so it doesn't super affect me. But man is it a bother to maintain.
It's interesting that the cost multiplier for Claude Sonnet 4/4.5/4.6 varies so much (1/6/9), while the API cost is exactly the same for all three models.
Also, the multiplier of 27 for Claude Opus 4.6/4. is way higher than the increase in API price would suggest.
I wonder why that is.
On GitHub copilot you pay per prompt. More powerful models can do a lot more work (consuming a lot more tokens) per prompt. Also, they tend to use more thinking tokens.
Why would folks be better paying 5.5% fee to OpenRouter ("Open") if most people just use one or two providers? Just use the provider's API.
One theory of the play of SpaceX might do if everyone migrates to query-based billing:
Provide cheap and unlimited access to Grok for programmers (hence the Cursor partnership/purchase for distribution).
-> This would drag massive revenue right before the IPO announcement, like if the company is super growing
-> At a loss, but don't worry, we need these funds to build the biggest datacenter of the universe.
This announcement would create enough momentum to increase valuation, and because of the merge of his companies, would save his X/Twitter investors from a tragedy.
-> Would also be a great service to Cursor investors and so, who are stuck with their VSCode fork
They probably want the training data. Otherwise these 60B don't make sense at all.
But they can't buy curser before their IPO so thats that?
Perhaps they have to much compute because Musk overpromised and Twittergroq doesn't need that much compute after he nerved the porn stuff?
It takes longer to build a datacenter with that much capacity than it does for the market to respond.
I don’t get the SpaceX reference. I thought they made rockets?
Nobody is paying for Elons xAI so he used SpaceX to buy xAI to fund it.
They now also own xAI
Which in turn owns Twitter. SpaceX is now a social media company in addition to a rocket company.
One theory I think Matt Levine posited, is that SpaceX will go public with dual-class stock that gives Elon control even with a minority ownership stake, and will subsequently buy Tesla, which doesn't have dual class stock, making SpaceX the singular "Elon Musk company", with him having operational control despite being public.
That theory aligns with Elon's long-held dream of X as the "Everything Company".
FYI, these are the multipliers for annual plan. I would hazard a guess most people are not on an annual plan
What's annoying is that it's obvious. In the case of GPT 5.5, if Copilot is going to charge 7.5x what GPT 5.4 costs while OpenAI themselves via the API/Codex only charges 2x of what GPT 5.4 costs, that will immediately raise an eyebrow.
To anybody who's been watching the tech sector with a critical eye for pretty much any period from the late 90s and onward, this is just the enshittification process. For most of OpenAI's existence it's been obvious, to me, that investors were burning insane levels of capital to build the market, and now that folks are locked in, you're seeing higher fees, ads, etc. Yet again, the user is the product; the investors want to siphon your data, attention and once you're hooked, money. And for companies like Microsoft and Apple, those hooks can dig deep.
I'd call it a straight up "bait and switch".
If you paid attention to the power requirements and amount of hardware being put into data centers, you should have realized that it cost them an order of magnitude more than you were being charged. To rework your analogy: they hooked you, now they're gonna see if they can reel you in.
They can only reel you in if its worth it. I still can code.
And while i do not spend 200$ privat, in my startup we discussed this and our current mental model is, that instead of hiring someone new, we prefer to have more money for tokens.
This is easier for us and has a bigger benefit. The cost of a new / first employee is very high, a 200$ subscription is not. Upgrading that to lets say 400 or 800$ is still alot easier and if i can run multiply and better agents with that money, lets goooo.
I'm looking at education -- teachers and students, not terribly tech savvy, are being mandated to use these tools. And then comes the rug-pull. It was worth it, but now it's outside of their budget. Poorer schools / students can't stay at the cutting edge; richer schools / students can.
Oh, I thought it was opium.
“Enshitification” is just when unsustainable subsidies end?
Another reason to hate that word.
From a different perspective, you were granted an incredible gift from the companies who let you use their product on their dime. Hopefully you made the most of it when you had the opportunity.
No, it's much more than that. It starts with unsustainable subsidies, as Uber undermined the taxi industry with a ludicrous burn rate. And then, once everybody's hooked to the point that they can't imagine life without the product, you raise costs. And you iterate: raising costs, lowering quality, selling data, increasing addictiveness. Until everybody wants to get rid of it, hates every aspect of it, but is still hooked to the core product. I'm personally not using these tools, not using uber or Meta products. But I'm still using some Google products and it's hard to extricate them from my life now that I'm using them.
We can't even get slop delivery worked out. So use SlopAggregator instead.
Everyone seems to believe OpenRouter isn't subsidizing but, until they publish audited financials, I personally doubt it.
OpenRouter doesn't even have hardware. What are they possibly subsidizing? The platform costs?
OpenRouter is guaranteed to be about the highest margin operator in the business right now. Everyone wishes they'd be them, skimming 5% off as the middleman without any OpEx.
Streaming, caching, and tool calling can get pretty expensive with scale, even when you don't touch inference. Maybe they're doing something clever and are quite profitable.. or maybe they've already taken $40mm from VCs and are currently trying to raise $120mm at a 1.3B evaluation.
They also show headline prices for the cheapest provider of whatever model, but then need to hit different backends some of which may be more expensive. For now they absorb those costs, but the VCs always come knocking.
Just my opinion though. Totally agreed that they have one of the best positions amongst all AI providers from a financial standpoint.
> They also show headline prices for the cheapest provider of whatever model, but then need to hit different backends some of which may be more expensive. For now they absorb those costs, [..]
They do?? I was under the impression I was just playing the price for whatever provider they deemed 'best' for each completion.
That is what I had heard.
Checking now: The way they describe it in their FAQ is that if the price changes, then they will bill you the new price. But I read that as regarding if the primary model provider changes their headline token cost; not in the case of pricing differences for models that have many different backends that host them.
Regardless, I would be more concerned about the streaming costs if the service continues to blow up and they scale aggressively through VC investments. If their 5.5% skim accounted for what they needed, you'd think they could effectively grow organically..
"eras" tend to not be so short lol
That's so unfair to us hard working developers. A month ago i could buy for .4$ a turn with Sonnet. Now i have to pay at least .9$ for this turn. Weeks ago i could buy for .12$ an Opus turn after they already raised prices and now they want .27$ from me for the same product! They are stealing from us!
They aren't stealing from us, for several reasons. First of all, it's a voluntary transaction. If you don't like the prices, use something else. Or don't use AI at all.
Second, you have no idea what their costs are. It is most likely that they are simply passing on their costs to you. If that was not the setup, users would just go to another service provider who was providing tokens at a cheaper rate. It's not like there is a dearth of competitors in this business.
The already stole when they trained their models on the data.
Now they just increase the price to buy it back
"Your plan pricing is unchanged: Copilot Pro remains $10/month and Pro+ remains $39/month, and each includes $10 and $39 in monthly AI Credits, respectively."
If there's no discount on credits (in terms of tokens per dollar) over other providers, I'm going to switch to a PAYG provider. If there's a month where there's little to no coding I can pocket the 10$. What incentive do they give to stay with this plan?
Yep.
Or if you're a business with multiple seats, these plans may be more inefficient than raw API usage billing. Since if anyone at your organization fails to utilize their full $19/39 allotment each month, that's wasting money, whereas with API credits it is 100% utilized.
I don't think they've thought through the implications of this. Everyone should cancel and go usage-based billing with caps.
They do address this in the doc, Orgs can now (although it was vague as to whether it was an option or just the new standard, probably option due to business contracts) 'pool' the Usage billing across all users.
I'm guessing they did that (and the 'temporary bonus credits') to make the pill easier to swallow for that side of customers.
You're right, I missed that.
It still does make one wonder, why have seats at all though? If everyone is just in one big API credit pool - what do the seats/users accomplish?
It forces you to pay at least $20 in tokens per user even for people who use less (they probably have stats on how many people use just autocomplete, which doesn’t count against the quota. or have a seat and don’t use the service at all).
For orgs, each user was allotted their own quota. For messages beyond that quota, a pooled budget is available.
They mention in the announcement that it will be possible to pool usage across an organization.
This was my first thought too. "Oh cool, I should be seeing lower prices" as I don't use Co-pilot that often anymore. But no, that's not the case. It rather served to remind me that I should probably just cancel.
They could add rollover balances and be back to cell phone plans in the early 2000s.
Are you thinking something like rollover plans?
Everybody who says it's a 5-9-27x seems to not be aware of the obvious loophole. More like 50x increase. You were able to use over $500 worth of Opus on a $10/mo Github plan easily, no hacks. You could just prompt "plan this out for me, don't stop until fully planned, don't ask any questions", and you would get ~$5 worth of planning in one 3x request. At 100 requests/mo, each easily reaching $5, that's easy $500 worth of tokens.
Even more so, questions and user answers from agents were not charged as separate requests.
And when you make your harness ask you for next steps in a tool call, the journey continues forever, yeehaa
This was my solution to very very though compiler tests that would take sometime up to 4 hours to figure out. Some of the time would be spent on running the tests, but still... I was burning so much tokens. I have free Copilot for my open source work so I wasn't even paying the $20.
this is the project that I am working on https://github.com/mohsen1/tsz
Bingo. I created a few autonomous skills that did exactly that for plan review, implementation, and branch review, review autonomously until green.
I was using 100M+ tokens per day, $250 per day or so and only paying $160 per month to GitHub.
I cancelled my GHCP sub and switched to Codex last week, so far so good but I miss Gemini 3.1 Pro for UI work.
Yeah it was crazy. Nowadays I use pi with OpenAI GPT 5.4/5.5, which to me seems both better and more generous than Claude. I supplement it with OpenCode Zen to get access to a bunch of models at token cost, and OpenCode Go ($10/mo) to get subscription-style access to Kimi, GLM and friends.
What is pi?
The minimalistic harness famous for having a tiny system prompt (avoids context pollution), and being what underlies openclaw. (https://pi.dev)
I did many 1h+ sessions of agent asking questions, delegating to subagents - all for 1 premium request.
I would say its a x1000 increase in price for agentic workflows.
They're not the only ones in the AI sphere to wind back, but they're the weirdest case in my eyes. Microsoft invests in having engineers building open models and they don't use a single one. I really don't get it.
But what really surprised me most about Copilot is that it would bill you per question, nothing about tokens. So if I managed to produce a prompt that gave me back an insane amount of tokens for something, which using any Claude model would easily accomplish, you were giving me my money's worth, at your own expense. The math is not gonna math out forever.
> I really don't get it.
One of the largest employers publicly engaging in a project which has the outcome of depressing wages. It's easier to "get" if you don't take the trillion dollar gorilla at face value.
"Plan prices aren’t changing.”
Isn't this like saying "The Porsche you rented at $200/mo is now a Honda. But the price hasn't changed!"
This may be a more accurate analogy... "The Porsche you rented at $200/mo now only allows you a maximum of 100km of travel. You will be automatically charged extra when you go over that."
A whopping 100km per month for the low price of $199.99!*
* with a quota of 138 meters per hour, overage charges may apply
more like 100m
On top of being worth less, the subscriber discounts are gone.
The old plans were $0.033/request for Pro, $0.026/request for Pro+ and $0.04/request for pay-as-you-go. That discount is now gone. They even still advertise "5x the number of requests" for Pro+ over Pro.
More like, The Porsche you had for a month you'll now have for 5 min only.
Yeah, if I go to a petrol station with 50€, but only get a tenth of the amount of petrol I got last week, I may think that the price has in fact changed.
"Your monthly fee isn't changing but it now only covers about 3 days of driving."
It’s technically true that the plan prices haven’t changed, it’s just the value you get from those plans has plummeted. It’s classic deceptive sales language.
More like, the rising gas prices aren’t a problem, I only ever fill up for $40
It's more like saying, "and you may now only use the Porsche for 5 minutes out of every day."
Full brake on the autobahn if you hit your 5min limit
They are now charging per gallon instead of a flat rate per trip
I was curious why a company would still use the VS Code + Copilot sidebar method for coding, rather than something like Claude Code. Turns out there’s a GitHub Copilot CLI!
I thought I was pretty familiar with available options, but no one in my circles ever mentions this product. It doesn’t seem to have much mindshare.
Has anyone used it? What’s your experience?
https://github.com/features/copilot/cli
I'm curious about the opposite: Why would anyone use the CLI when, at least with Copilot, the VSCode plugin is super tightly integrated with VSCode, meaning the agent can see everything I can see. There's no mismatch in linter calls where I can see a lint in the ide that the agent can't find for example. I've had this problem even using CC in their VSCode extension, so I can't imagine it's not an issue in the CLI as well.
What's actually better in the CLI?
We need sandboxing for any agent, so we run it within Docker - so we use CLI.
I use vscode with containers extensively. Not sure why containers imply CLI.
Yeah, I've been using it heavily at work since the beginning of January (and have a personal Anthropic sub to compare to). Copilot CLI is pretty good, honestly. Most new features in Claude Code get cloned by Copilot CLI within a couple weeks. Claude models seem mildly more clumsy in that harness than the one they're trained on - subjective guess around 20% more turns for an equivalent task - but it's not a noticeable difference in the final output.
The vs code integration is pretty slick. I can copy and paste function names into the prompt and it automatically turns them into these `#sym:` reference objects that I presume populate the context window with metadata about the function and where it lives. It knows what file I'm currently looking at as I jump around in the code, and that automatically gets loaded into the context. I can also drag and drop folders or specific files for context into the sidebar.
It's a lot of stuff that makes me have to type less into the prompt, since it's already getting so much info from my editor
It's in their ToS to allow using Copilot subscription with OpenCode - https://github.blog/changelog/2026-01-16-github-copilot-now-...
Absolutely the cheapest way to get a lot of tokens through a solid harness for $10/month. Until now
> I was curious why a company would still use the VS Code + Copilot sidebar method for coding, rather than something like Claude Code.
I use Claude Code, but I kept my Copilot subscription around mostly for really cheap usage of other models when I need to try a different one (which appears to be ending, in a sense) and also the autocomplete in Visual Studio Code which was really great across a bunch of files, I could make changes in one file and then just tab through some others.
I wonder what other good autocomplete is out there.
> also the autocomplete in Visual Studio Code which was really great across a bunch of files <...> I wonder what other good autocomplete is out there.
I am in the same boat. I tried looking for tab/auto-complete implementations ~ a year ago and it was pretty disappointing. If that has changed, would love to know!
I've used it quite a bit. There are a lot of AI terminal coding products and this is another one. It works well, handles sub-agents without issue and does a reasonable job operating in the Copilot ecosystem. It handles mid-task questions and such we well.
I’ve tried OpenCode, Claude Code and Codex CLI. But was just shocked that Microsoft has a version I hadn’t even heard of.
Personally I got CLI fatigue and am happy with Conductor for now, but things are moving fast in this space.
The naming is bad. VS Code Copilot Chat.
But its a really good UI for agentic coding. Not sure why more people don't use it. I've tried the others and keep coming back to Copilot chat. It's a really good tool. Which is why the rugpull on pricing is so concerning.
I’m actually trying to move back from the Claude Code style, I feel like it’s easy to become distant from your own code, and I am feeling uncomfortable with that.
I’ve “vibe-coded” some projects and when I start to find issues or go to refactor them I don’t have that memory of why decisions were made, because many decisions were never made.
The other cool thing is Copilot SDK, so you can build agentic capabilities into apps, or build tools, that leverage the agent harness of the Copilot CLI:
https://github.com/github/copilot-sdk/
I'm just so confused why people aren't just using ghostty/kitty/terminal.app and claude code. Compared to the other approaches I've tried, it's by far the most effective way to get performance from opus 4.6/4.7
I don't know about others, but I use Copilot more often than other apps because of its tight integration with the VS Code itself where I still spend most of my time working on other things while letting AI do some task that I decided to delegate to it.
That's espoused as the big reason for the price increase: most Copilot subscribed developers it seems have moved to "agentic usage" with the CLI and Cloud-based agents.
Which feels a bit like a kick in the pants for me as a developer that was primarily using Copilot for VS Code ghost text and very rarely used the Chat sidebar much less "agentic" tools.
Copilot Pro sort of made sense for my personal account when amortized across a year, but I don't want to "waste" $10/month on credits I won't use most months.
I've been using it for a few months because Copilot was the only AI blessed by our corporate overlords. It's not bad, I would say it's about 80% as capable as Claude Code, which I've used extensively on personal projects. However CC was recently approved, and I'm betting that with these changes to Copilot pricing we'll end up dropping it like a hot potato.
I've used it. It's on par with OpenCode imho.
I tried the VS Code + Copilot sidebar approach a few months ago. It was definitely rough around the edges compared to Cursor/Claude. In our corporate environment, we weren't even able to use frontier models.
it's my favourite harness so far
Because Copilot is the only thing allowed at our corp
Search has become so bad that I also struggled to find Claude Code alternative and made my own tight (not editors, not plugins, not agents, strictly similar to Claude Code CLI) list: https://github.com/omarabid/cli-llm-coding
The list is not long but there are quite a few options. Even Grok has its own CLI!
The reality is, even though a CLI prompt looks very simple, it's a very complex piece of software. I personally use Claude Code (with GLM) and anything else I have tried was significantly inferior (with the exception of opencode).
Windsurf made a similar change in March: https://docs.windsurf.com/windsurf/accounts/quota
> In March 2026, Windsurf replaced the credit-based system with a quota-based usage system. Instead of buying and spending credits, your plan now includes a daily and weekly usage allowance that refreshes automatically.
With hindsight, per-request pricing makes no sense at all if an agent can burn a widely varying amount of tokens satisfying that request. These pricing plans were designed before coding agents changed the dynamics of token usage.
I wouldn't call it hindsight - I don't think anyone, at any stage, thought running a 10 minute+ sonnet session for 1 premium credit was ever profitable. We all knew it was a loss leader to get people using it.
It would have been profitable if that premium credit cost more than a negotiated discounted rate with Anthropic. We have no way of knowing if there were negotiated rates though!
There is no way to make that cost model profitable consistently. If 1 prompt can mean 100's/1000's of requests over hours, and you only pay for that 1 premium prompt, that can never be profitable.
Guys, you're discussing a house of cards to begin with: No matter how you're paying for the $CURRENTSOTA you're not garunteed that next month what you pay for will be the same.
So, lets do some honest evaluations:
1. The model itself is a non-deterministic engine of work with an unknown value; it's real value is just magic.
2. The business model itself is non-deterministic engine of profit with a known value; whatever the VCs have put into it, _must_ be piulled out. If Ed Zitron's numbers are correct, circa 2030, it's several trillion dollars.
So do some matrix multiplication of non-determinism vs determinism, and realize that the value proposition for _you_ is only going to decrease because #1 can never outpace #2, ensuring enshittification captures a smaller and smaller whale.
We know this. This has been the last 2 decades of money extraction from software. It was ok when it was some 12 year old's parents CC. But now it's you, or your business, that's going to either ben squeeze for value or squeeze out of the market.
And everyones squabbling about the color of the cost. ok
The problem with assuming that tokens can only get more expensive is that the Chinese open weight LLM firms have dropped models which have a known, fixed price that can never get more expensive (since we can run them on hardware we own).
per-request was broken, yeah. but $10 of monthly credits is basically just a prepaid wallet with a reset timer.
I wonder if GitHub (Microsoft) is implicitly betting that enterprise demand is sticky enough to absorb these rates, especially given that Opus 4.6 “fast” was being listed at a 27x multiplier. Maybe they saw enough usage at that price point to conclude the demand is real. Or maybe the strategy is to keep the enterprise customers who can justify it while shedding heavier individual and power-user usage.
The interesting question is how long it takes enterprises to notice the capability/pricing tradeoff, and whether they respond by limiting access to the strongest models internally.
The part that worries me is that this market is still very early. Most developers and organizations are still learning how to use these tools effectively. Raising the experimentation cost this much may slow down the discovery process that makes the tools valuable in the first place.
I also suspect that there are many "slow-moving", Microsoft heavy enterprises but with in-house devs that can't get anything but Copilot approved, and Microsoft trusts this will remain so.
It's not turning consumption based because there are a ton of these licenses just sitting idle.
As a single data point, this is absolutely true. At my current "Big Corp", Copilot was immediately approved while Claude is entering month 2 or 3 of trying to get approval.
Additionally, we got copilot for every user, including those that never write code or use AI tools.
Subsidies stop when LLMs improvement plateaued (though they still benchmark higher somehow). At some point, you have to make money or at least break even; and I think they concluded that we reached that point.
As someone that is on the enterprise side in a non-tech F500 company, what I'm seeing is some FOMO and need to be part of the hype cycle. We're about to plonk a bunch of money on more Copilot licenses. Something got in the water where all the C-levels the past two months are pushing everyone to use AI but when they bring up examples of their uses its like "I use it to rewrite my emails" or prompt 'engineering' ideas that point more to patching over poor processes, data management, and decision-making within the organization or not.
What we're seeing across the board is every software company tossing AI onto their name or sales pitch and no one understanding what that actually means. But we will spend money on it because of FOMO.
I really question if we're reaching the end of the hype cycle to the point. I wish I were brave enough to put money on it. It feels like there was a command from up top to 'do something with AI' and leadership is scambling for some resume-building projects vs doing the hard work they should've done the past two years at a people and process level.
I really hope that they think so and that they're wrong and they get burned hard. Them and all the AI labs that lied, stole, inflated, hoarded and tried to justify all this as an existential moment where AGI would radically change society. I hope their calculus to reel in paying users is all wrong and now they all crash and burn instead of recouping VC money.
I was always curious how they can sustain a request based pricing model when requests can range from tiny to huge with all the modalities GH Copilot offers. Was a steal for agentic coding that turned out to be too good to be true, in the end. Still: Thank you for the ride.
I don't use Copilot or any paid AI but all of this usage-based billing reminds me of cellphones back when you paid per individual text message.
Usage paying for AI is 1000x crazier because you're not even getting a guarantee in the thing you pay for in the end. You have to keep feeding it prompts and hope it gives you the solution you want. You may end up with no expected result yet you are paying for it. At least with texting, you got what you paid for.
I wonder how long it'll be before all AI costs are flat unlimited monthly fees or even free across the board, without compromise.
It incentivizes you to do most of that prompting on your own hardware/time, and only feed the final prompt with only necessary context to the big AI in the sky. It might even force you to think about the problems yourself for a bit!
I expect in the future we'll find out that someone in the industry was juicing the numbers with fake thinking tokens or something. The whole pricing model of charging you for the tokens it generates while not knowing how much it is going to generate going in has always been pretty crazy.
Yeah, this was my frustration with Suno and Sora. You can burn a lot of credits (not to mention time) generating things that aren't what you wanted.
I don't mind a PAYG model for a simple chat interface. But when it comes to actually producing things, you burn through TONS of tokens creating the wrong output.
> I wonder how long it'll be before all AI costs are flat unlimited monthly fees or even free across the board, without compromise.
That's already the case if you can self-host an LLM; you don't even need a mythical H200: gamer-grade GeForce cards can get you a long way there (if this page is to be believed: https://www.runpod.io/gpu-compare/rtx-5090-vs-h200 )
...after RAM prices return to normalcy, of course - and then wait another 2 or 3 generations of GPU development for a 96GB HBM card to hit the streets - and also assuming SotA or cloud-only LLMs don't experience lifestyle-inflation, but I assume they must, because OpenAI/Anthropic/Etc's business-model depends on people paying them to access them, so it's in their interests to make it as difficult as possible to run them locally.
Give it 5 years from now and reassess.
Github had, by far, the most easily game-able agent usage policy. People would force the agent to run a script before the end of turns that consisted entirely of `input("prompt: ")` so that you could essentially talk endlessly to an agent for the price of a turn. I see this less about the future of this industry and more about fighting the costs incurred by bad actors.
I never played any games like that, but simply giving the agent a clear exit criteria and instructions to check the exit criteria every time it thinks it's done on a complex task was often enough to keep it chugging away for most of a day on a single prompt in my experience. Per-prompt pricing just isn't sustainable period, even if everyone is acting in good faith.
Charging by prompt was always wild to me.
I once asked it to do a comprehensive security review of our code. It churned for nearly an hour (and then produced 90% false positives). Insane that that usage was charged the same amount as me just saying "Hello".
There is noticeable trend across all agentic coding platforms that this situation is no longer sustainable.
With this kind of pricing (sonnet 4.6 has 9x multiplier, previously 1x) it begs the question why use Copilot to begin with.
You could easily just buy the tokens directly and have a lot more choice as well.
One reason I used it was that I wasn't locked into a single provider and switching them was as easy as changing a drop-down. Small feature? Sonnet or GPT5.4/mini? Large changes? Opus. And why not see how good Raptor Mini does this one refactor?
It also helped build an intuition of what wach model could do and which parts it was weaker at because you could try them almost side by side, especially if one model's output wasn't great.
That said, these were all side projects so nothing truly consequential. Otoh, you might leave some extra perf on the table but I found the models worked quite with the Copilot harness.
Yeah, this is a very useful abstraction layer. The entire concept of separating the model creator from the model runner is good for competition and is customer friendly. Which means they likely hate the concept and want to kill it.
Gosh, imagine getting to do that with your TV/Streaming subscription. Getting to pay one fee to access some set number of hours per month from any of the providers.
Doesn't GitHub get volume discounting they can pass on to their Copilot customers?
Economics of scale don't work when scale still isn't enough and capacity is still limited.
GitHub has the full power of Azure with their hosted models but it's not being passed to consumers.
Economics of scale don't scale
It seems to me more expensive but I might be reading it wrong.
Looking at their pricing it does not look the case.
The problem is I can't afford the tokens! Even on my $10/mo plan, running either 100 opus, or 300 sonnet agent runs would cost hundreds of dollars - well above my budget!
I started to use github copilot with vscode, but have never been too happy about the system. Over the months I gravitated to much more agentic workstyle, hardly ever editing much code by hand. The vscode IDE was getting more in the way. I had already started to look at OpenCode, and when I found it has a web interface, I was happy to switch over. I use a simple editor (KDE's Kate), or just less to skim through the code and/or a git diff. OpenCode has some free models in it, but I think I will need to get some kind of subscription for a better one. But it won't be copilot any more. The market is moving so fast that I don't know what are the most resonable models, or the most flexible way to set them up so I can switch when prices change yet again.
Check out Zed sometime, its pretty decent too.
So given that I primarily interact with LLM's through VSCode, and I prefer the Copilot interface to the Claude Code plugin, does anyone have any suggestions on other plugins I should try? In my experience, Copilot is much more "plugged in" than any of the other plugins, in the sense that it can see things like linter outputs in VSCode. Basically, copilot "sees what I see" in a way that no other plugin or command line tool can, which make it much more ergonomic to use.
With this pricing change, I see no reason at all to stick with Copilot in principle, but I really need to solve this issue of IDE integration to move on.
You can use Copilot Chat* with basically any API provider, and if you switch to the VS Code Insiders build you can configure it to use literally any OpenAI API-compatible endpoint.
Other than that Zed has a similar experience which is pretty decent.
* By which I mean the good one, whatever it's called now - the part of Copilot that used to be a plugin and is now part of VS Code, not the thing that has always been part of VS Code.
I used to feel exactly like you do, but now I use the Claude CLI exclusively and am very pleased with the results.
Could you elaborate on what's better about the CLI?
Try Cline, a vscode plugin that's been around since the start and has really active development. You can also use most providers as well.
Saw this new Cline fork mentioned in another post: https://github.com/dirac-run/dirac
Also heard of more and more people moving to Kilo Code or OpenChamber instead.
What's the current situation for coding with Local LLM's on decent hardware? I have an M3 Max with 64 gb of ram and am thinking I should start looking at Ollama and Opencode? Is this a useful stack for smaller personal projects?
Use llama.cpp or better yet Unsloth Studio
It’s getting there. You could give a try with qwen 3.6. It’s worth paying for better models in the cloud, but local models are now better than nothing.
One nice development recently was ollama's support for MLX optimization on Mac hardware. It's not obvious how to know you're using a model that works with it, yet, so it's rough around the edges.
https://ollama.com/blog/mlx
After seeing the ridicolous multiplier increase I've added a calendar event to cancel my subscription mid-May.
(I'm a copilot subscriber since 2022)
Already cancelled here! One less thing tied to GitHub.
Has anyone found the answer to this yet?
> What is the benefit of using the Copilot Pro+ at 39$/month instead of using the Copilot Pro at 10$/month and paying for extra usage?
Some models, for example Opus 4.7 and GPT 5.5, are only available on Pro+; Pro+ has audit logs and GitHub Spark; that's about it, as far as I can tell from https://docs.github.com/en/enterprise-cloud@latest/copilot/g...
If I had to guess...
On my personal account, Copilot Pro+ still only gave me back Opus 4.7, whereas my work's Pro account still lets me use Opus 4.6.
So, my gut says, it's entirely possible that Pro+ will continue to have more segregation on model availability...
FTA
> Last week, we also rolled out temporary changes to Copilot Individual plans, including Free, Pro, Pro+, and Student, and paused self-serve Copilot Business plan purchases. These were reliability and performance measures as we prepare for the broader transition to usage-based billing. We will loosen usage limits once usage-based billing is in effect.
There's enough weasel wording here that I would expect only certain models get re-enabled on Pro.
e.x. lots of people seem to get good enough results from Opus 4.6, personally I prefer it over 4.7 in GH Copilot... locking that down to Pro+ would be, given this salvo of enshittification, a 'logical' move on their part.
The cheapest copilot plan felt totally unsustainable to me. For around £8 month i was getting 100 opus 4.6 prompts (albeit with a reduced context window size around 128k iirc vs 200k to 1m for first party hosted opus). Gpt5.4 was hosted with 400k context iirc.
On top of that, you’ve got 2000minutes of container runtime, so running cloud agents was included. As was anthropic agent sdk mode via copilot which is very comparable with claude code - not identical, the anthropic “modular prompt” is much leaner in the sdk version.
I cant say im mad, i got above what i paid in value. That said, going forward ill probably go back to openrouter payg rather than a subscription.
I got a free 3months of the gemini £19 plan and ive been playing quite a bit, 3.1 pro is a good model, i just find it slow. Flash i think i under appreciated until now.
I pay for Copilot annually, and mostly for its code auto completion features. I use CC if I want to do anything agentic. Not sure if I want to pay more for occasionally-good-intellisense at this point.
Same! I wonder what other alternatives there might be for autocomplete.
If you find out, please let me know!
autocomplete is still unlimited within the subscription, maybe using a free model or even cheaper ones are best value..
Code completions and Next Edit Suggestions remain free with Copilot Pro.
But you can no longer amortize annually, which makes it even more a question of "is this worth it this month?" each month. Especially for personal accounts.
I'm similarly thinking about sticking with the auto-downgrade back to Copilot Free when the annual sub ends and then just yelling about it any months I hit the 2000 completion cap.
I wouldn't mind a plan between Free and Pro that is just "all I care about is code completion and next edit suggestions".
I liked copilot because I didn't have to think about tokens. I get hung up when having to think about the price of things, and its hard to think about the project at the same time I got to think about token usage like a gas bill. The usage system had its own issues, but having a set amount of requests was a very comfortable way to use a paid AI service.
Sounds like you're a candidate for a local model. It's kinda nice not caring what the token count means except as to compaction.
Not paying per token? Not sending my code to someone else's servers for inference? That's the stuff of sweet dreams for a stingy, paranoid solopreneur like me.
If I could run a local model comparable to even Sonnet 4.6 without shelling out $50K in hardware, I'd do it in a heartbeat. But all I have is a 32 GB of RAM and an old RTX 4080.
Or am I not up to speed? Are there decent coding models that can run on dev laptops? Not that that's what you were suggesting by recommending a local model, necessarily; just curious.
I do love using local models when I can, but qwen-35B is the best model I can run, and while its an insanely good local model, it does not compare to the big ones.
Have you tried the latest Gemma? You might prefer it to Qwen, depending on what you're doing.
I did, but in almost everything I tried even qwen3.5 was better, and 3.6 was a huge step up.
Looks like Microsoft has run out of compute and can't scale it fast enough to serve copilot users and Azure AI Foundry needs, given that the customer base is growing there as well.
vindication! https://www.wheresyoured.at/exclusive-microsoft-moving-all-g...
I thought of this as soon as I saw the headline: I hope this is the start. Crossing my fingers to see more line up with your predictions.
Current multipliers vs from June
EDIT: only applies to annual plansI think that only applies to held-over users on the annual plan:
> Users on annual Pro or Pro+ plans will remain on their existing plan with premium request-based pricing until their plan expires, however, model multipliers will increase on June 1 (see table).
Kind of. Monthly users moving to an entire different pricing model so we don't really know what the increase in price will be for them.
There will surely be corresponding, different AI credit costs for each model.
"AI credits" are just US cents. The cost per model is here - and it looks like it is just the API providers API cost: https://docs.github.com/en/copilot/reference/copilot-billing...
Thanks this is helpful. I am on my first month of annual Pro. Should I keep it or cancel it (ask for a refund)?
It isn't just the big multiplier increase, they also say "...and no new models or features will be added to annual plans going forward."
Can you imagine ten months from now and you're still rolling Sonnet 4.6?
Cancel/refund is looking pretty good. They're doing refunds until May 20.
"To request a refund, go to Settings → Billing and licensing → Licensing, select Manage subscription, then choose Cancel and refund "subscription". (The phrasing varies slightly depending on your subscription ). This option will be available until May 20."
Appreciate the info. Have you got a similar opinion of the 'convert to monthly' option they've also provided?
Not apples for apples.
Before:
- Opus 4.6 each premium request is 3 premium requests
After:
- Opus 4.6 each dollar spent is 27 dollars in copilot AI Credits.
Given that you'll receive 19 dollars of AI Credits in Business plan, that means you can probably say 1 "hi" to opus per month.
It is an apples for apples comparison since those new multipliers only count if you are on an annual plan in which case the premium request system stays in place until you either cancel and get a refund or until your renewal comes up. https://docs.github.com/en/copilot/concepts/billing/usage-ba...
If you are not on an annual plan, multipliers will be gone completely. You can see the rates that apply instead here: https://docs.github.com/en/copilot/reference/copilot-billing...
Thanks for that link. Oddly the blog post didn't contain any of this information.
Based on the pricing and comparing to competitors e.g. bedrock[1] looks like cache-write will only be on 5 minute TTL.
[1]: https://aws.amazon.com/bedrock/pricing/
GPT 5.4mini is even worse, from 0.33x to 6x which is ~18 times more expensive now.
GPT-5.4 and GPT-5.4 mini are now the same price, which, why?
Those multipliers will only apply if you are currently on an annual subscription (and only until your renewal comes up or you cancel). So I assume they simply want to make it as unattractive as possible to get most people to cancel it and move to the token based system.
That's not an answer. It's specifically a discrepancy between 5.4 and 5.4-mini. If you look at all other models/generations you see that the cheaper model indeed has a lower multiplier. It's very strange that only 5.4 doesn't have this.
both 5.4 were best bang (-mini even more as I found it usually same well performing!!!) for the buck before. Now they face cost reality it seems.
GPT 4.1 had a multiplier of 0.
I was surprised to find that this sentence
> Plan prices aren’t changing
did not continue with an em-dash followed by something profound that is changing.
Plan prices aren't changing -- the value you get out of it is.
How is this legal when people paid for a yearly plan in advance?
In order to most-to-least charitable, any of:
1. Github could choose to grandfather in those plans and make no changes until those plans expire.
2. Github could offer, or the user could request, a pro-rated refund along with cancellation of the account.
3. Tough luck, those users agreed that Github could unilaterally change the ToS at any time.
Right now there is a cancel and refund button on the Github Copilot annual subscription setting, which I have just pressed.
> 1. Github could choose to grandfather in those plans and make no changes until those plans expire.
They explicitly stated that they won't be doing that: the multipliers go into effect in June for everyone, annual plan or not.
In places with reasonable consumer protections (Australia, Germany) it almost certainly is illegal unless they give a full (whole year) refund. I think the short time limit of applying for a refund won't be looked at favorably either. Regardless of their ToS which I'm sure covers this.
But companies do lots of illegal things, and in general nobody takes them to court over it.
I doubt you can force them to provide the service with the original terms, but you might be able to ask for a (partial) refund. If not today, after a week of verbal abuse they will receive for this online.
It depends where you’re located. In the EU they have to honor the contract you entered, but presumably there is a clause that they can prematurely terminate the contract without cause and give you all of your money back (from the start of the contract).
100% it says in their terms that they can change the service during the agreement.
That kind of clause would be void in many places around the world.
For example, the German Civil Code states:
Let’s see if German users can enforce it or not.
My thought exactly! First the usage limits + model limitations and now fundamental change to the billing. Hope some consumer watchdogs are looking into this!
I just checked and you can cancel with a refund.
For the yearly plan they only change the model multiplier. And it's in the subscription contract they can change that multiplier at any time.
I guess this is the "Google App engine" of Vibe coding when google raised pricing of App Engine significantly after being in preview. Doing weird price changes like this when coming out for new services make much more sense than doing this change which will just anger users.
What's the residual value of Copilot after the changes? For Enterprise plans, even copilot code reviews will be charged at token price + github action minutes for the execution of the review. One can roll one's own reviewer and have it spend the API tokens as well. At least one will be able to select the subset of files for the change set, if needed.
The background agents will also depreciate in value because of their harness that's a black box that's not optimized for token usage at all. Rolling one's own will be a better choice here.
so what's everybody using to get autocompletions in vscode? i've been using copilot just because $10 is cheap, but i use opencode for everything other than completions.
i tried the continue vscode extension, and it seemed kind of janky. are there better options?
I bought a copilot subscription for some small personal projects at Christmas.
I haven't been able to use my subscription much over the busy spring months, but i'm being charged every month.
I'd be tempted to keep the subscription if usage-based billing meant that i'd save money when i had less time.
But today, after hearing this, i cancelled my subscription.
Does this mean you can only prompt "Hello" every morning for a month with Opus 4.7 ?
Well, the free launch for somebody else's money is over.
So I guess from now on GH Copilot is only worth it if you want a quality autocomplete in VSCode.
That was the first thing I turned off in VSCode. Autocomplete for my TypeScript projects was great. And the "AI" suggestions/completions were really getting in the way of me still being the "driver."
I'd be fine with it if they can make Sonnet 4.5 unlimited. I haven't personally seen any major differences between 4.6 or any of the other newer models. Claude 4.5 seems to be the right balance and works great for me.
"Paid for annual? Tough luck, from now on your usage limits are reduced by 89%. You can do 11% of what you paid for. Good luck if you paid annual a month ago!"
And then they have the gall to say
> "The bottom line: Plan prices aren’t changing"
If anyone lives in a place like Germany or Australia and has an annual sub, please take them to court, you're guaranteed to win because you have reasonable consumer protections and their ToS doesn't stand a chance. 9x reduction is unreasonable and the consumer cannot be expected to see this coming.
In case some diehard enshittifier believes that consumers should know better and businesses should be allowed to get away with it, where is the line? 99% reduction? Is that still okay?
If this situation is to be acceptable then it should be regulated as a financial product like stocks, which come with knowledge tests of "do you know you can lose all of your money?". And come with regulatory compliance and all that.
According to `bunx ccusage` I'm easily doing $250-400/day in "real" API costs on my $200/month plan. There's no way everybody else isn't going to do the same thing and completely change the industry again. Both beginner and advanced developers are already hooked on all this stuff and they all know it.
what's the best place to transition to on a company level after this change ? should companies just switch to pay as you go with open router ?
It begins.
"It" being the end of subsidization of tokens and plans (expected) but while lock-in to foundational models and cloud services is still lacking. Guess investors want their ROI sooner than later, given how big of a wrench the AI boom has thrown into global economics.
In light of this, does anyone have a DGX Spark and use it as a coding agent?
Built credit pricing into my SaaS for AI features and the hardest part wasn't the math, it was that customers can't easily predict their own usage. They underuse and feel cheated, or overuse and churn. Subscriptions hide that volatility from the customer. Usage based pricing makes it their problem, which is honest but harder to sell.
We all knew this from the very beginning but couldn't compete with OpenAI or Anthropic on their subscription-based pricing strategy. It was nuts except for those few corporations burning investor money to keep competition out as long as possible. Now they don't have to hide anymore that subscription pricing won't do it for ai. The pyramid scheme is falling.
The problem is that people expect to get the output of 100 people with a $20 subscription by spawning multiple agents. This is unrealistic. I'm using 2 codex plus account and able to manage a repo with 265-300k lines of code.
The point is - if its the same or more expensive per month than a real human employee - why pay for AI ?
Human retain knowledge, product knowledge, can pick up more work often for the same money. And having many of them means your business wont go down if provider suddenly bumps API pricing.
How much is the average pay for a junior developer in US? Its definitely much costlier considering PF and other benefits. Just the math. if you use it efficiently it's much cheaper than hiring a permanent staff. You can maintain a lean team and do all the mundane boilerplate coding with AI.
some of Github's open source maintainers have lost their free github copilot pro, guess this is really the next step for them to save cost in their infrastructure.
Here goes my Copilot Pro subscription then, reluctantly heading over to Codex CLI since the CC base plan is downright unusable.
from these comments I'm going to make mouse.dev pricing as follows
-BYOK runs are $0 to Mouse, period.
-Hosted runs are billed at provider cost + a published markup.
-We will never invent a unit of billing that isn't denominated in tokens, seconds, or tool calls.
-Credits in the paid category never expire.
Whose idea was this “premium request” model anyway? If you’re going to invent a new metric used to bill, why not align it with what, even at the time, was a clear underlying cost structure that GitHub actively chose to ignore for a more confusing system.
It made more sense in the ye old days where a request was basically just a chat message in a sidebar and it could also edit code. Then saying someone can use 300 chat messages a month kinda makes sense.
Turns out when a request can spawn tens of subagents and use millions of tokens over many turns of toolcalls then suddenly github copilot has a massive financial problem on their hands.
This approach started with the “Ask a question about your code” feature, which is more comparable to single chat message with relatively predictable token usage. Now it’s an agent who might work for 30 minutes, read the whole codebase, and write 1000 lines
30 minutes lol. I've gotten Opus to work for a full 8 hour day on one prompt.
I'm not usually a Conspiracy Guy, and the answer is probably `incompetence * tech_debt`. But I think that having sufficient layers of abstraction to any billing model is a useful way to hide the real cost of things. It's why it's done everywhere.
In business, if a strategy is making more money, the most direct and usually correct answer is malice, not incompetence.
As a Github Copilot user, who mostly just uses chat in the VS Code editor but still burns through my Pro limit every month -- what's the best alternative price to performance? Claude Code?
I keep hearing that Codex is the best bang for your buck now.
Just got an email with this announcement.
I have Copilot Pro that I use occasionally, but not enough to tell how the switch to per use would affect my usage.
Based on description Pro plan users will get $10 in monthly AI Credits, but that seems rather low compared to what you could use same plan until now.
> rather low compared to what you could use same plan until now.
That's exactly where the subsidy is being removed.
Soon it will be cheaper to just do it yourself
It may already be cheaper if you calculate the hidden cost accumulated by LLM generated tech debt. At least some of it could be avoided by manually doing it better to begin with.
This reminds me, I have to cancel co-pilot...
It was just a matter of time, considering how many Mtok you could consume in just 300 prompts.
Why would anyone stay on the Pro+ plan going forward? Pro with openrouter for Opus would be cheaper?
So what's the best alternative now? Openrouter + cline or something else?
i like opencode's zen https://opencode.ai/en/zen with opencode's TUI.
End of an era for predictable costs as a small business. We will refer to these times as ‘the good old days’.
I expect the prices to rise to almost cover the cost savings from layoffs at large companies.
cursor, windsurf, and CC are all already on usage-based models so I guess what really matters is whether Copilot's GitHub integration depth justifies the price per token vs the alternatives
haven't touched copilot for one year, this reminds me to cancel, it will take sometime to catch up
And so it begins...
I'm happy I invested in local solutions and cutting context to the bone for API providers. Claims about AI being able to fully replace programmers never took into account the long-run equilibrium price of inference.
Already there are companies paying more for coding tokens than for programmer salaries.
That's my point. They made those decisions without any consideration for the long run, which would have required them to project the cost of AI services years into the future. Obviously management didn't do that and had no way to do that. It made current earnings look good, though, which was enough for them when they made the decision.
Not sure what your point is, but they are firing lower performing programmers so that the top programmers have more tokens to spend.
Doesn't necessarily mean that's a good idea.
I'm not sure I understand this. All I know is now, I pay $39/month (actually less because I paid a year up front), use the agent, mostly on auto--and only choosing a model if it got stuck or in a loop--every day, and haven't hit any limits yet. It seemed to good to be true, after hearing others talk of $300/month bills. I guess it was.
This subsidized inference is just a marketing ploy to increase prices and profit.
If common people can have a DIY setup with an open source model cheaper than those behemoths with a scale advantage, it's clear that we have been played.
Time to either self host a Chinese open source model or to just pay the cheap Chinese providers.
Yeah, local is clearly the future. Even beyond the cheap Chinese models you can install the apfel[1] stuff if you're on a mac and want a quick available onboard cli option. And I'm sure people will adapt the Flash-MoE[2] integration to be even better soon as well.
[1] https://apfel.franzai.com/ [2] https://github.com/danveloper/flash-moe
I really don't understand why OpenAI, Anthropic and Microsoft are in competition to see which one of the three will elevate deepseek the most.
DeepSeek will do the same thing.
Z/Mimo already raised their prices multiple times since the promotional prices at the start of the year.
DeepSeek (and other open weight models) can be served by anyone with the means to do it. All of the big open weight models are available on OpenRouter via commodity providers.
DeepSeek is actually more expensive than expected ATM due to compute shortage, but they said it would come down in price.
The compute shortage will be bigger 3 months from now.
Glad this was announced because I didn't even realize that our (small) team has been paying $20/month for Github Copilot when none of us are using it, so used the opportunity to cancel CoPilot altogether. I think it was free when I first activated it (this was also before Claude, Codex, Gemini, and while not that great, it was why not), and didn't realize it had switched to paid and bundled with our Github bill, and now per usage.
So about a month left before cancelling. Got it.
... Once again the Business accounts get all sorts of goodwill [0] and users get the shaft.
[0] - Last weeks changes limited my personal Copilot Pro account but not my Work one
What "goodwill"? It's just more "AI credits" for what will be a shit product in June.
Which one is it:
1. Current models in fact do not solve coding.
2. You can simply wait for a ~year for open-source to catch up and run it locally.
Re 1: Current models don't solve coding. They are useful tool for it though.
Re 2: Open weight models seem to be less than a year behind proprietary ones, so sure, if you're willing to spend tens or hundreds of thousands of dollars on a super computer that you probably don't fully utilize instead of renting time on someone else's super computer for a lot less.
Github Copilot has for the last week been an absolute shitshow. Clearly something has happened behind the scenes.
AI itself clearly weaker / dumber. Prices increased manyfold.
"The bottom line" is the new "Its not just X, its Y"
aaaand another github outage today. FFS
People need to wake up and stop being surprised by these billing increases. I see it on every update of every model. This was all subsidized by VC and company money. Now they need a return and the prices will keep going up. Be glad that you took advantage of that up until now, but can we stop the pearl clutching when we all know the amount of money being dumped into AI and the lackluster returns?
It's less surprise, but more confusing given the game theory as their competitors are not doing the same thing and the multiplier changes alone will likely churn current users.
AI companies are increasing prices across the board.
TLDR: It's a 6-9x price increase
tldr: people were running multi-hour agentic coding sessions for the same flat fee as a one-liner autocomplete, github was eating the bill, and that party's over on june 1st
Still no chips coming from Microslop
Google won