GitHub Community
I used up my entire month's credit limit in just one day. #197571
Replies: 37 comments 54 replies
-
|
The individual responsible for the pricing strategy should be reassigned to another role, as their approach is extremely short-sighted. |
Beta Was this translation helpful? Give feedback.
-
|
maybe you can use deepseek V4 flash. Adding this model to copilot using DeepSeek V4 for Copilot Chat plugin |
Beta Was this translation helpful? Give feedback.
-
I used deepseek v4 pro, Subscribing to Copilot is no longer worthwhile. |
Beta Was this translation helpful? Give feedback.
-
|
I am starting to use claude code, I think I will not use the copilot if they do not adjust their price strategy |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
|
copilot pricing is terrible. This is June 3rd 2026 and i'm already 53% in on AI credit. that's bad. Really bad |
Beta Was this translation helpful? Give feedback.
This comment was marked as spam.
This comment was marked as spam.
-
|
I used agent mode, My project isn't large, but my credit limit is being used up too quickly, so I've decided to cancel my subscription. |
Beta Was this translation helpful? Give feedback.
This comment was marked as spam.
This comment was marked as spam.
-
|
I paid the bill, but they cut off my PRO Subscriptions and the help team does not exist! |
Beta Was this translation helpful? Give feedback.
-
|
I think that its time , we need to move to other than this one. |
Beta Was this translation helpful? Give feedback.
-
|
Same here. Under the old plan, $10 easily covered my monthly usage. With the new point system, I'm burning through the daily limit much faster while doing the same work. At this rate, the cost difference is huge. |
Beta Was this translation helpful? Give feedback.
-
I have already subscribed to cursor |
Beta Was this translation helpful? Give feedback.
-
|
just throwing my comment in the mix that my $39 plan was completely used up in about 2 hours of normal use. I will not let my subscription renew if there isn't a change to this. i'll be finding another provider. i heard codex is good these days, may try that. i used claude previously and didn't hate it. also, i have never before posted on github community before today; never felt the need to, but this change was so bad i felt the need to come and speak out about it. github, microsoft, i hope this is seen and understood and rectified. i get they were trying to balance things, but they balanced things in the same way anakin skywalker brought balance to the force |
Beta Was this translation helpful? Give feedback.
-
|
Same. I've never felt the need to post or follow here. Already cancelled my Pro+ subscription. Ludicrous |
Beta Was this translation helpful? Give feedback.
-
|
I too am burning through my premium credits. I used ask, not agent, for one question about css, and it cost me 4% of my usage. Does anyone know of a good alternative to github copilot |
Beta Was this translation helpful? Give feedback.
-
|
I will do the same I will cancel my subscription after 4 years, I was one of the first users :( |
Beta Was this translation helpful? Give feedback.
-
|
My case is even worse. A single use to refactor a type in a class used 70% of credits... On a 1500 rows .py file for like 50 lines of code... On a single request. Basically, after 3 minutes I cannot work anymore for the whole month and must migrate somewhere else... They must re-design their approach |
Beta Was this translation helpful? Give feedback.
-
|
That sounds painful. If a $39 plan disappears in two hours of normal agent usage, the real problem is not only the price — it is that the cost is hard to control before the agent starts running. For coding-agent workflows, I’d look for an alternative setup that gives you:
A cheaper model is only useful if it reduces the total cost of a completed task. If it causes more retries, the savings disappear quickly. I’m working on a China AI API model hub for global builders — one API to access the models independent developers and AI teams need. If you’re testing another provider, I can share a small free test credit so you can compare cost, latency, streaming stability, and accepted output quality on a real coding workflow. |
Beta Was this translation helpful? Give feedback.
-
|
I wasn't surprised and expected to burn through my full month in one day, same thing happened with anthropic in may. Let's see how accurate these valuations are as they hemorrhage subs by the thousands from these new, unjustifiably high, unregulated rates and fuck them for it. |
Beta Was this translation helpful? Give feedback.
-
|
Last month, I upgraded from the Pro plan to the Pro+ and was happy with the usage; it was perfect for me. As of today, when my credits reset, with a few requests, my usage is already at 13% and continues to rise even with very few requests. I've opened a ticket and hope they'll respond soon. |
Beta Was this translation helpful? Give feedback.
-
|
Me too |
Beta Was this translation helpful? Give feedback.
-
|
Yup, I couldn't believe it. Rug pull at it's finest! |
Beta Was this translation helpful? Give feedback.
-
|
Damn this new price and mechanism is crazy.. |
Beta Was this translation helpful? Give feedback.
-
|
Minimax coding plan and olllama both integrate into the vs shell and minimax just dropped m3 |
Beta Was this translation helpful? Give feedback.
-
cursor or deepseek v4 pro |
Beta Was this translation helpful? Give feedback.
-
|
I feel this. For coding agents, “best value for money” is not only about the cheapest token price. The better metric is usually cost per accepted task: how much it costs to get one working refactor, bug fix, or implementation merged without too many retries. I’d compare providers by:
DeepSeek V4 Pro / Flash can be a strong cost-performance direction, but I’d test it on one real coding workflow first instead of judging by one prompt. I’m building a China AI API model hub for global builders — one API to access the models independent developers, AI SaaS teams, and automation teams need. If you’re comparing Copilot alternatives, I’d be happy to share a small free test credit so you can benchmark cost, latency, streaming stability, and output quality on a real coding task before committing to another provider. |
Beta Was this translation helpful? Give feedback.
-
|
One day? You’re a lucky guy. Some of them hit 100% in under an hour |
Beta Was this translation helpful? Give feedback.
-
|
Haha you are right |
Beta Was this translation helpful? Give feedback.
-
I was already shocked when I asked the first question, so I immediately switched to DeepSeek V4 Pro, and only used GPT-5.4 for difficult problems. But in just one day, the quota was used up like that. |
Beta Was this translation helpful? Give feedback.
-
|
THIS IS SOOO BADDDDDDDDDDD! ALL CREDITS GONE IN JUST ONE PROMT ! |
Beta Was this translation helpful? Give feedback.
-
|
That's just comedy. Cancel your GitHub Copilot subscription and just buy the Claude Code. So much cheaper compared to raw API pricing if you are using Anthropic models. After the recent changes to the GC pricing, all you are doing is pre-paying for API rates, which is really REALLY bad deal. You are getting the ugly price of the API while also risking losing your prepaid tokens if you don't use it in the month. You are certainly better off just buying the API key from the provider directly and wiring it in yourself. But it is much smarter to buy Claude Code or the $20 Codex sub if you use OpenAI predominately. For me, I cancelled my GC Pro+ subscription last month and have been using Cursor Pro ($20) plus Claude Code Pro ($20) for the same price. I think it's a very nice combo. With Claude Code you can continuously use agentic work with Sonnet for 3-4 hours every 5 hours, or about 1 to 1.5 hours every five hours with Opus 4.8. No nosebleed pricing from Microsoft with their arbitrary multipliers and their totally hidden session and week limits. It's a really great product for most developer's needs. And Cursor's Composer 2.5 along with many other models like GPT 5.5, Kimi, Grok, etc fill the gap for when Claude Code hits their limit or when you want to try a different model. |
Beta Was this translation helpful? Give feedback.
-
|
and please how much does that cursor cost for 20€? is it for tokens like with API or are there any hourly limits? Thanks for the answer |
Beta Was this translation helpful? Give feedback.
-
|
Cursor uses two separate pools: the Auto+Composer pool with a very cheap rate that doesn't count towards $20, it is almost free. And it uses $20 for API pricing, so I use that sparingly only when Claude Code times out or when I want to use a different model. So my recommendation is Claude Code or the Codex for primary work, and Cursor for a nice IDE and extra models and Composer 2.5. |
Beta Was this translation helpful? Give feedback.
-
|
Remember to turn off 'Fast' in Composer 2.5 settings by clicking the 'Edit' tab (still really fast at a much reduced inference cost). By default Composer 2.5 defaults to Fast (nearly 5x the cost). This simple change will significantly decrease usage consumption but you'll get equal quality output. |
Beta Was this translation helpful? Give feedback.
-
|
Cancelled after several months (51 company subscriptions). Absolutely ridiculous. |
Beta Was this translation helpful? Give feedback.
-
|
Same here PRO+ usage usually last me heavy workloads 2-3 Weeks. Now a SINGLE day of work 100% USAGE OF TOKEN, will need to switch i guess. |
Beta Was this translation helpful? Give feedback.
-
leave from Copilot |
Beta Was this translation helpful? Give feedback.
-
|
As of June 1, 2026: Usage-based billing for GitHub Copilot is now live for all users and Copilot code review consumes GitHub Actions minutes, in addition to GitHub AI Credits. As part of this release, we're also launching new user-level budget controls, expanded context windows, and enabling upgrades to Copilot Max. ⭐ For the most up-to-date information, we have provided more details and a FAQ in this discussion. |
Beta Was this translation helpful? Give feedback.
-
|
I think we all noticed.... Hence so many of us unsubscribing. Although only a tiny insignificant amount of money we have cancelled all of our subscriptions as of 1st June 2026. That is some 51 subscriptions going elsewhere... Whomever made this decision at GitHub needs to be fired as I'd estimate in one day alone they'll probably have lost you a couple of million Dollars and counting. Thank you for what was a great service up until today. Hitting 100% in just over an hour is an absolute insult. |
Beta Was this translation helpful? Give feedback.
-
|
This is an extremely bad decision. I've never though of cancelling my subscription, but this is just a terrible decision from your end. I'd definitely do that. |
Beta Was this translation helpful? Give feedback.
-
|
I'm seriously considering canceling too. In less than 1 day I have already spent 60% of my monthly tokens. |
Beta Was this translation helpful? Give feedback.
-
|
I have the same experience. I burned through 80% of my Pro+ credits in an 8 hour session. and after another 2 hours I was consuming my overage budget at a fast rate. I just upgraded to Pro+ a few days ago as the weekly, session and other new limits they bestowed upon us prevented me from even using AI Chat at all. I received the 5 hr limit warning for the 3rd time in 24 hours, So I went to bed and about 8-10 hours later I logged in and sure enough, I was beyond the 5 hour session limit, 8+ hours after the fact. The huge multipliers really suck, they are up-charging for everything, it a huge cash vacuum. But what's worse is the new dropdown to select the model no longer shows the Free models at the bottom and the premium models on top, it doesn't show any multipliers. And the Models in the IDE don't come anywhere close to matching the supposed Models available listed on the web site. There is no way I will continue to pay these prices, at my current rate it would cost me $1500 monthly if I used Claude Sonnet 4.5, |
Beta Was this translation helpful? Give feedback.
-
|
canceled |
Beta Was this translation helpful? Give feedback.
-
|
I requested my data, then cancel and delete account! |
Beta Was this translation helpful? Give feedback.
-
|
Perhaps it's a mistake, my Pro subscription was used in a day and and that rate even if I upgrade to Max at 10X the cost it won't last the month. Feel like I've been scammed by a Ponzi scheme |
Beta Was this translation helpful? Give feedback.
-
I think so too. It feels like it would take at least 30 times as much to be possible. |
Beta Was this translation helpful? Give feedback.
-
|
Monthly I was using 20-30% of my limit, I didn't need more, I'm a good coder without assist ;P but yesterday my monthly usage was eaten by only one prompt. |
Beta Was this translation helpful? Give feedback.
-
I've started using DeepSeek V4 Pro. Even though it's not as good as GPT-5.4, I have no choice — it's cheap enough. |
Beta Was this translation helpful? Give feedback.
-
|
Hello everyone, how are you? No joke, I made just 3 prompts and it used up all 200 credits for the entire month. That's absurd. |
Beta Was this translation helpful? Give feedback.
-
hahaha, 😄 |
Beta Was this translation helpful? Give feedback.
-
|
I've also just cancelled, used all my pro credit in about 3 hours, that was with an additional $5 credit on top |
Beta Was this translation helpful? Give feedback.
-
|
I was getting my company of 200 people ready to sign up after the poc. We will switch to another provider now, this has no longer anything to do with honest business but robbery instead. Sayonara |
Beta Was this translation helpful? Give feedback.
-
I used to worry about AI replacing humans, but now it seems that the cost of AI is not low, and it also brings the uncertainty of black-box decision-making. |
Beta Was this translation helpful? Give feedback.
-
That is a very important signal. If a 200-person company was ready to register after a POC but is now switching to another provider, then this is not just a “developers dislike pricing” problem. It becomes a procurement and rollout-risk problem. For company-wide AI coding adoption, the key question is not only whether the model is good. It is whether the cost can be predicted and explained before rollout. For a team POC, I would compare providers by:
A provider can look reasonable for one developer, but become very risky when scaled to 50, 100, or 200 seats. I’m building a China AI API model hub for global builders — one API to access the models independent developers and AI teams need. For a company POC like this, I think the safest test would be a small controlled benchmark: one or two real coding workflows, non-sensitive code only, clear cost/latency/output notes, and no full migration until the numbers make sense. |
Beta Was this translation helpful? Give feedback.
-
|
Hello, We have provided more details and a FAQ about usage-based billing in this discussion, which may be updated if new information becomes available. To keep conversations all in one place and to ensure you receive timely updates, we’ll be closing this post and recommend you subscribe to the discussion linked above. |
Beta Was this translation helpful? Give feedback.
-
|
What terrible PR! Those are not the questions being frequently asked, not even remotely. The top FAQ should be "WTF????", followed up with "What other service should I switch to?" How disconnected from your actual customers could you possibly be? Just read the discussion you linked, and the discussion here. You're not answering the right questions. I guess on a larger scale view, this could be a good thing... it's proof that AI in it's current state isn't sustainable... and we can all get back to a normal work pace! |
Beta Was this translation helpful? Give feedback.
-
|
I will also skip using vs code and Github copilot and move to cursor, claude or windsurf. Microsoft is robbing us! |
Beta Was this translation helpful? Give feedback.
-
|
Sadly, I think they all will do this to us in the end. |
Beta Was this translation helpful? Give feedback.
-
|
I used like 4 prompts and 50% credits are already used up. Unacceptable. |
Beta Was this translation helpful? Give feedback.
-
|
leave Copilot |
Beta Was this translation helpful? Give feedback.
-
|
It is the 4th of June and I've used all 3200 included credits for my organisation, and 9500 additional credits. Currently on around $100. Let me put it into perspective, I used to pay $75 monthly, since AI credits: 4 days into the month I'm on $95. I've not increased the usage at all! I thought maybe $100 a month is semi-acceptable for it to do maybe 10% of my job, not $750+. Not to mention this is a organisation account (simply because of the transition period, we could not use Opus unless you were an org, our projects require decent intelligence), and I've single-handled used all the credits within 3-4 days - If my entire org was using this, we'd be looking at ten's of thousands each month, we barely even get paid that per project and Copilot will do maybe 10% of what is necessary! (There's a lot more to projects than just coding!) $750 could be over 50% of a persons wage (particularly part-time) in some places, you think this is acceptable?! |
Beta Was this translation helpful? Give feedback.
-
|
I'm currently using Claude Code DeepSeek V4 Pro. I'm considering using the advanced models of Cursor, Windsurf, and Codex as supplements.Microsoft must be crazy.No one would use such an expensive AI. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks @DAAworld I will be taking a look into these! |
Beta Was this translation helpful? Give feedback.
-
|
This is now week 3 since I first contacted GitHub Support. The budget caps I'd previously set on my organisation don't appear to apply to the new credit-based Copilot billing SKU. I wasn't aware of that at all, and I ended up blowing through what was previously an $80–90 monthly spend in just 4 days because I expected my existing budget limits to actually... limit the budget. What made it even more frustrating was that I stopped using Copilot completely, disabled everything, removed every seat from the organisation, and the usage charges still kept going up. In the end I deleted the organisation entirely just to be safe. Three weeks later I've still had absolutely no response to the billing dispute, despite chasing it twice. That's the bit I find unacceptable. Billing disputes shouldn't sit unanswered for weeks when it's people's money. Anyway, I've moved over to Claude Code directly on the Max 5x plan with the VS Code extension. There was a bit of setup getting my skills, agents and memories sorted again, but apart from that it's been a really smooth switch. Honestly, Sonnet feels way better through Claude Code than it ever did through GitHub Copilot. I can't explain why, but after using both for a good while it genuinely feels like a different experience. I still use Opus occasionally when Sonnet isn't quite enough, but that's pretty rare now. If you're mainly using GitHub Copilot for Claude models, I'd seriously consider giving Claude Code a go instead. The pricing is predictable, I know what I'm paying every month, I haven't hit the Max 5x limits yet, and even if you do the allowance resets every five hours. There is a weekly limit too however again I don't think I've hit that yet and it also seems to oddly replenish slightly overnight. For context, I work on some pretty large and complex projects every day, so I'm not exactly a light user. If this plan is enough for me, I'd imagine the $19 plan would probably be enough for a lot of people too. Between the billing mess and how much better Claude Code has been for me, I just can't recommend |
Beta Was this translation helpful? Give feedback.
-
|
You know I was looking around I can buy a mini PC with 2400 dollars, and run my own AI, and I will not pay for them for 2 years at least ! |
Beta Was this translation helpful? Give feedback.
-
|
This is ultimately what I think is what the market will land on once high VRAM gpu solutions become more accessible. AI shouldn't be THIS centralized, anyway, imo! |
Beta Was this translation helpful? Give feedback.
-
|
Yes I've been working on my own coding assistant AI myself hoping to get away from the centralized systems. |
Beta Was this translation helpful? Give feedback.
-
How? |
Beta Was this translation helpful? Give feedback.
-
|
I build my own FREE alternative to Copilot, with Ollama local, and some free models and working GREAT, even from Telegram, I can WORK now, This move from Github, make US better to move ON! |
Beta Was this translation helpful? Give feedback.
-
By using Python and my own library in combination of ROCm TheRock for Native Windows. You can also use llama.cpp. |
Beta Was this translation helpful? Give feedback.
-
|
I feel this. The painful part is not only the new price, but that agent-style coding makes the cost hard to predict before the task starts.
DeepSeek V4 Pro / Flash can be a good direction for cost-performance, but the real test is one real coding workflow, not just a single prompt. I’m building a China AI API model hub for global builders — one API to access the models independent developers, AI SaaS teams, and automation teams need. If you’re comparing Copilot alternatives, I’d be happy to share a small free test credit so you can benchmark cost, latency, streaming stability, and output quality on a real coding task before committing to another provider. |
Beta Was this translation helpful? Give feedback.
-
|
Did Microsoft say anything about changing or going back to the way it was before? I didn't find any news saying if they spoke out. |
Beta Was this translation helpful? Give feedback.
-
|
I already ended my subscription after 3 years with GitHub copilot , I am using Claude now and its working fine with the 20$ plan. |
Beta Was this translation helpful? Give feedback.
-
|
Just had a training from a guy at Microsoft on Github Copilot. The guy was just mocking us by using the most expensive model available all the time on high and max context window, he stated he had unlimited credits. He burned up our company budget several times during the training. Funny enough he did not succeed in making the app he so overconfidently started to make work =D. |
Beta Was this translation helpful? Give feedback.
-
|
That training example is exactly the kind of rollout risk I was talking about. Using the most expensive model with max context all the time may look impressive in a demo, but for a company it creates a very different problem: the budget can disappear before the workflow proves real value. The painful part is that the app did not even succeed after burning the budget. That means the real metric should not be “which model was used,” but:
For a company POC, I would avoid starting with the most expensive model and max context by default. A safer test is controlled routing: cheaper/faster models for routine steps, stronger models only when needed, and a clear budget cap per workflow. Otherwise the demo can look powerful, but finance and management will see only unpredictable spend. |
Beta Was this translation helpful? Give feedback.
-
|
Or, you know, run a Local AI on-premise and skip that headache entirely. |
Beta Was this translation helpful? Give feedback.
-
That is a fair point. For some companies, local or on-premise AI is absolutely the right answer, especially when privacy, data control, and predictable infrastructure spend matter more than access to the newest hosted models. The tradeoff is that “local” does not remove cost completely. It shifts the cost from API billing to hardware, GPU availability, maintenance, model updates, concurrency limits, latency, monitoring, and reliability. For a company POC, I think the best comparison is not cloud vs local in general, but:
Local AI can be a great part of the stack. The key is measuring whether it actually completes the workflow more reliably and predictably than hosted options. |
Beta Was this translation helpful? Give feedback.
and others
Uh oh!
There was an error while loading. Please reload this page.
-
🏷️ Discussion Type
Question
💬 Feature/Topic Area
Copilot in GitHub
Body
I used to get 300 premium requests for $10, and by the end of the month I'd have 100 left. Today I got 1500 points a day, and I've already used them all. At this rate, I'll need to spend almost $200, or even more, to finish the original job.
Beta Was this translation helpful? Give feedback.
All reactions