New token system and scaling is 10X trash

[TheStorkClub]

🏷️ Discussion Type

Bug

💬 Feature/Topic Area

Copilot in GitHub

Body

Its 56 minutes as of the writing of this and using a copilot pro+ subscription and on some relatively simple html work using GPT5.4 XH (due to the lower cost) - I am already at 13% of my monthly usage..... what has gone on? this is ridiculous scaling of usage where something that would be 10% for an entire day of heavy use... whomever in the business thought that this scaling was a good idea just lost some business....

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐

[marchlqq10]

I’m extremely disappointed with Copilot Pro+’s new billing model.
In the past, a regular question with GPT-5.5 used only 1% of my monthly credits, around $0.40, even when I let the AI run without interruption. Now, just one simple query that lasted 3 minutes — and I manually stopped it halfway — ate up 5% of my total quota, costing roughly $3.3. The cost has increased by at least 20 times.
What’s worse, if I work on complex tasks that take 1 to 5 hours to process, my entire $39 monthly credit allowance could be exhausted by a single question. In such cases, the cost surge would range from 10 times to as high as 10,000 times compared to before. This billing scheme is completely unreasonable and makes the service practically unusable. I sincerely hope the official team will investigate and fix this severe billing anomaly.

[lamp76]

I've Just burned all the AI Credits of my GitHub Copilot Pro Plan, 1500 AI credits gone in a few hours!

I'm using a GitHub Copilot agent, Gpt-5.4, and in a few sessions just burned all the montly Plan credits and Is using also the additional budget.

I hope the there is a bug in the new AI crediti attribution, otherwise I'll cancel my subscription immediatelly.
I think that GitHub is trying me to pay the time consumed by the errore on their bad code, look at the errore, thousands of repeating errori for a lot of minutes.

Can help me to contact the GitHub Copilot support?
Ok for the new AI Credit model, but I refuse to pay for they syatem errors!

[maheerCodes]

I can understand the frustration, but it may help to gather a bit more information before concluding that the new credit system is behaving incorrectly.

A few questions:

• Approximately how many prompts or requests were made during that 56-minute period?
• Were you using GPT-5.4 XH exclusively, or did you switch between models?
• Were the conversations long-running with large context windows?
• Were you generating large code blocks, refactors, or multiple file edits?
• Does the usage dashboard show any unusual spikes or breakdowns?

Without those details, it's difficult to determine whether this is a pricing/scaling issue, a reporting issue, or simply higher-than-expected token consumption due to context size.

If you can share some rough usage metrics or screenshots (with any sensitive information removed), the community may be able to help identify what is driving the credit usage.

If you later determine that the usage is expected and your question has been answered, please consider marking the most helpful reply as the answer so others can benefit from the discussion.

[TheStorkClub]

Its also an account of how much this new plan has throttled the token usage of the system to the point of arguing why should I keep subscribing for usage that will run out in a week...

[TheStorkClub]

It would be awesome to provide you my usage from earlier today but it looks like your finance and product team also thought it was a great idea to throttle this usage too.... The great squeeze is upon us...

[maheerCodes]

Thanks for sharing the screenshot.

From what I can see, it confirms that you've already exceeded the included credits, but it doesn't show where those credits were consumed or provide a comparison against the pre-refresh usage pattern.

The part that stands out to me is your comparison between:

• ~5% usage over a full day before the change, and
• ~13–16% usage in roughly an hour for what you describe as similar work.

If those numbers are accurate and the workload was genuinely comparable, then I agree that's a significant difference and worth GitHub investigating.

The challenge is that the screenshot doesn't expose the per-request or per-conversation breakdown needed to determine whether this is:

1. a change in credit accounting,
2. a model pricing change,
3. context accumulation from long-running chats, or
4. an actual reporting issue.

At this point, I think clarification from the GitHub/Copilot team would be the most useful next step, because users need to understand whether this level of consumption is expected under the new system or not.

[TroyCoderBoy]

@TheStorkClub Get rid of Copilot and subscribe to CODEX or Claude with a MAX plan. Way better value.

[markdicks]

I've only prompted 6 times today and I'm on 15%. It's insane. Normally I'd only be 15% in after 3 days or so. It's been 1 hour

[cozuya]

Yeah I'd buckle up. Just showed up here because the new system you put in 1.5 hours ago looks like this:

All of my monthly credits were used by one code review prompt and didn't finish on something that would have been done with 1 premium request last month, 1/300th of our monthly meter.

[cozuya]

gpt-5.4 xtra high on a pr

[cozuya]

ITS NOT EVEN JUNE YET

[CarmJos]

Fxxk, me too. I just use copilot to review my code, after that, I cannot use anymore.

[maheerCodes]

That's a useful detail.

If the request was a PR review using GPT-5.4 Xtra High, then the model tier itself may be an important part of the explanation. Large pull requests can generate substantial context, and review workflows often involve significantly more processing than a typical chat prompt.

That said, the concern being raised here isn't just that usage increased—it's the apparent magnitude of the increase compared to what users experienced before the rollout.

At this point we have at least two reports describing a much larger-than-expected jump in credit consumption after the transition. Whether that's due to revised pricing/accounting, model-specific costs, or something else isn't clear from the available data.

Hopefully GitHub can provide guidance on what level of credit consumption users should expect for common workflows such as PR reviews, code generation, and long-running conversations under the new system.

[TheStorkClub]

@maheerCodes - your openclaw is still running on this - you gotta be careful of your own tokens you might run out as well!

[maheerCodes]

😄 Fair point.

At this rate I may end up becoming part of the dataset.

Jokes aside, I do think the interesting question now is whether what you and @cozuya are seeing is expected under the new credit model or whether the rollout introduced a much larger effective cost increase than users anticipated.

Hopefully someone from the Copilot team can clarify that, because multiple reports describing similar behavior are starting to emerge.

[FurtherAway-Dream]

“That's ridiculous. I used to be able to ask 1,500 questions a month, but now I can only ask about 130 or even fewer.”

[TheStorkClub]

I'm literally giving up on prompts that are costing too many tokens and doing it manually instead .... the Great AI transition.... Already up to 20% of my monthly token usage....

[czinczaar]

the new system is crazy greedy. I made a small agent request on a script with 500 lines with gpt 5.4. The request didn't require extravagant reasoning, but just adding a little bit of functionality. It added 41 lines and removed 6 lines. But it already cost 6% of the 1500 AI credits(so 101 credits).So it means that if I make 10 more small chat agentic request like this one, I would already be at 60% of the 1500 AI credits? Are you serious? This is absolutely ridiculous. I wanted to wait for the new system to arrive before I make a decision but this is retarded. I am cancelling right now.

[bettyposteringwilly-netizen]

I CANNOT believe how expensive this got all of a sudden. SO disappointing.

[lionng429]

I got this problem as well. One simple prompt request and it burnt 330 credits in minutes.

I immediately set the additional usage budget to $0 to avoid potential billing shock. It is truly disappointing, and now I'm going to switch to Claude.

[lionng429]

FYI, I just ran another similar request with Claude Sonnet 4.6 model, but switched to Low Thinking Effort. It burnt 150 credits. So the $10 Pro plan would only do 10 requests for me in a month.

[czinczaar]

I really hope this is a mistake on their part and they will fix it, otherwise they absolutely lost their minds

[moonlune]

Yes, the token costs under the new mechanism are unbelievable! Changing the explicit cost multiplier to a hidden graph is, in itself, a deliberate attempt to obscure hidden fees! I tested “Claude Sonnet 4.6” and “Gemini 3 Flash (Preview)” and found that, more often than not, their token consumption is not much different! This is completely at odds with the official promise that “Gemini 3 Flash (Preview)” would consume roughly the same amount of tokens as the original version!
Perhaps this approach does indeed allow the platform to turn a profit, but it will also drive users to other service providers! Of course, maybe that’s exactly what the platform wants?

[CarmJos]

W.T.F? I can only run 5 AI commands with current AI credits.

Completely unusable.

R U KIDDING ME COPILOT?

[tlingard]

On the Pro+ plan woke up this morning, May 31st, with somewhere around ~35% request used for the month (~65% remaining -- pretty typical usage for me). Set a long running plan execution on auto pilot this evening and grabbed some food. Come back to check progress and by 9pm on May 31st, received:
402 {"error":{"message":"You have exceeded your monthly quota","code":"quota_exceeded"}}
-- Resets on Jun 30 at 7pm

(yep, checked billing setting, and it already switched over to June usage)

that's just ... awesome copilot. It's not even June yet.
To quote one of my all time favorite github comments: "what in the actual name of fuck"

[zeuszwc-afk]

Feedback on Copilot points deduction: As shown in the screenshot, I only submitted 2 queries, yet nearly 10% of my points were consumed. Even the highest-tier plan priced at 100 USD does not support many queries.

[bettyposteringwilly-netizen]

I use VSCode. What are my options for using another AI coding assistant? Can anybody recommend one for me?

[TroyCoderBoy]

I terminated my Copilot Pro+ subscription and subscribed to a Claude Max plan and love it. I would recommend subscribing to either a OpenAI Codex plan or Claude Max plan.

[esthrim]

I'm planning to subscribe two accounts of claude pro subscription, which cost $40, same with the github copilot pro+, and can be cheaper only $34 per month if annually. But the minus is we need to switching between account. but it's more worthed.

[pharazone]

39 dollars usage worse than 20 dollars codex plus! You are so Avaricious!

[loto12345]

It is exactly 10x more expensive than before.

[santosshrimps-jpg]

Where is my legal documentation ?? As today no body have come to my address with any legal paperwork to address any criminal charges or restrictions!! You want to demand ?? I am waiting for you at my house bring documents and introduce yourself properly and then me and you we can talk !! Otherwise you are just an email!! Best regards. Santos Get Outlook for iOS<https://aka.ms/o0ukef>

…

________________________________ From: loto12345 ***@***.***> Sent: Sunday, May 31, 2026 10:30:36 PM To: community/community ***@***.***> Cc: Subscribed ***@***.***> Subject: Re: [community/community] New token system and scaling is 10X trash (Discussion #197524) It is exactly 10x more expensive than before. — Reply to this email directly, view it on GitHub<#197524?email_source=notifications&email_token=CDUPGZ6C2NFCXFDZA2KFA5T45UIHZA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZRGMYDIMBWUZZGKYLTN5XKU43VMJZWG4TJMJSWJJLFOZSW45FMMZXW65DFOJPWG3DJMNVQ#discussioncomment-17130406>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/CDUPGZ33P7NC2ZTVQNL4GJ345UIHZAVCNFSM6AAAAACZU3CQXKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTOMJTGA2DANQ>. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS<https://github.com/notifications/mobile/ios/CDUPGZYI5MG53NTRJXOCV5L45UIHZA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZRGMYDIMBWUZZGKYLTN5XKU43VMJZWG4TJMJSWJJLFOZSW45FKMZXW65DFOJPWS33T> and Android<https://github.com/notifications/mobile/android/CDUPGZ5L2JCF5FIFJZI2ZU345UIHZA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZRGMYDIMBWUZZGKYLTN5XKU43VMJZWG4TJMJSWJJLFOZSW45FOMZXW65DFOJPWC3TEOJXWSZA>. Download it today! You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Arinluna]

Besides the fact that I'm probably once again cheaper than the AI... There is another problem with the increased price, much harder to control by GitHub.
I'm fairly early in the development of an app, so when I noticed that some of my tabs don't show as they should, I requested an app frame level change, so this problem is solved for the future for all possible tabs. I was prepared that this is not going to be cheap. But the AI decided last minute that it should revert the framework level change and do an in place patch, for one of the affected tabs... (All in the same session, I had no chance to save the work.) If your human employee does this, they get disciplinary action, not payment.
33% of my monthly credits wasted in about 56 minutes, on the quick but bad solution that would take me 5 minutes to do.

[marchlqq10]

I feel the same. Just two or three basic questions have cost me 1000 credits already. Conservatively speaking, I can only ask about 21 questions in total. For complicated queries, I’d get no more than 15 uses. My team and I are planning to switch to other AI services.

Totally agree. I paused my first question this morning as soon as I saw credits soaring. I only had three queries in total — two in the morning and one in the afternoon — and my usage is already at 1013 credits. I won’t risk using it further.
A colleague of mine has used 32% of his quota (total 7000). He will run out entirely within one or two days. Spending $39 for just two days of access is absurd. It’s no longer just wasting money; it’s literally tossing money down the drain.
Conservatively, I can only ask about 21 simple questions, or 15 complex ones. My team and I are preparing to switch to other AI solutions.

[gurtner-it]

Same for me... One prompt only, 500 credits used. crazy. so i could only do 14 of those, and it was a rather simple prompt...

[dw7086]

需要放弃pro+，10倍消耗，这不是改计价规则，这是错误

[LElfvin]

Hopefully they refund, replenish, or otherwise fix it today. Releasing a new system with a 10-20x increased cost is insanity. Not the first time this year this has happened either... Sent 6 prompts before I glanced at the new token usage system, over 34$ usage in 15 minutes is crazy!

[Fluidios]

Same here. A 30-minute session—and 25% of my limit is already gone... It's time to look for other code agent solutions.

[awrff]

My token consumption is ON FIRE LOL. Can not believe ONLY 3 requests cost 49% of my business subscription!

Honestly, I don't think users will learn how to cut the usage since the increment is insane. It's way easier to just jump to another subscription.

[Fluidios]

Maybe that's their plan.

[awrff]

In that case, I'd give them a thumbs up, getting rid of users indeed saves the money lmao.

[HienHT2003]

All of my credits were exhausted by a single prompt, incurring an Additional usage charge of $5.72 USD. Fortunately, I terminated the session. I do not understand what Copilot was processing in the background. I am highly disappointed.

[sHyShcA]

The most expensive AI tool I’ve ever seen. I hope this is some kind of mistake on GitHub’s side and they’ll refund the money.

[TakSeBiegam]

for minor latex doc changes, mostly done on gpt-5-mini + caveman, that's insane how much tokens you gave for ppl.

[sathish-mygit]

seriously, bill scales more than 20x now. I think this some scaling up for some bigger pockets

[ChaoyiLiu-neusoft]

Copilot is heading towards its demise; this is a premeditated massacre!