Update to GitHub Copilot Consumptive Billing Experience

[Akash1134]

Hello Copilot Community,

We’re excited to announce an update to your GitHub Copilot experience: monthly premium request allowances are now in effect for all paid Copilot plans! This change is designed to give you more transparency and control over your premium Copilot usage.

What’s changed?

The monthly allowance of premium requests per user is now enforced for Copilot Pro, Pro+, Business, and Enterprise plans. Premium requests unlock a selection of advanced AI models and features—usage varies by model.

What’s unchanged?

All paid plans still include unlimited use of GPT-4.1 and GPT-4o for agent mode and chat interactions, plus unlimited code completions! Please note that rate limits may still apply across all models.

Managing Your Premium Requests

• Monthly Allowances: These reset on the 1st of each month. For this first billing cycle, your premium request counters have all started fresh at zero.
• Pay-Per-Request Option: Need more premium requests? You can set a spending limit in your billing settings (default limit is $0).
• Usage & Billing Controls: Monitor your real-time premium request usage via the Copilot status icon in your IDE, or download detailed usage reports in your billing settings. If you have licenses from multiple organizations, you can select which entity to bill.

For all the details, check out our documentation.

We Want Your Feedback!

Have questions, comments, or suggestions? Your input helps shape the future of Copilot, please comment below.

Thank you for being a part of our community and helping us make Copilot even better!

[iwangbowen]

300 premium requests are not enough for pro plan

[wrek]

May we suggest the pro+ plan?
And coming soon the premium pro+ plan... Then I'm sure we'll see the enhanced premium pro Plus plan...

[knoopx]

300 per day is ok, per month is ridiculous

[gadfly3173]

• Usage & Billing Controls: Monitor your real-time premium request usage via the Copilot status icon in your IDE, or download detailed usage reports in your billing settings. If you have licenses from multiple organizations, you can select which entity to bill.

The real-time premium request usage mentioned in the announcement is clearly still not available to all users. Why has this feature been delayed for so long?

[Akash1134]

Hi @gadfly3173 This could be a feature deployment delay, IDE restart might fix this.

[hayoshii]

It feels like instead of making Pro+ a better plan, you’ve simply made the Pro plan worse.
This change doesn’t enhance the value of higher tiers—it just limits existing functionality for paying users who expected more consistency in what they’re paying for.

[hayoshii]

This is my third year using Copilot, but to be honest, while more unnecessary services and features have been introduced, placing a 300-request cap feels quite unfair.

Since it doesn’t seem like this change will be rolled back, I’m considering switching to Cursor once my current annual Copilot subscription ends.
If you have any other recommendations, I’d be happy to hear them.

[raky291]

@hayoshii An alternative for me would be to use local models with Ollama + VSCode or Cursor.

[hayoshii]

@raky291 Thanks for your input!
Of course, I’ve been experimenting with local LLMs as well.
At first, I only intended to use them as a supplement to GitHub Copilot… but here we are.

[nijaru]

For API usage, I'd recommend the anthropic API directly for claude, otherwise openrouter is great for switching between whichever model you want to use.

If you want purely subscription based, not API billing, then claude code with a claude pro or higher sub, or else go local with something like ollama.

[hayoshii]

@nijaru Thanks! Claude Pro is definitely on my list of options, but OpenRouter sounds especially interesting.

[returnvoid]

Nothing exciting about this change. Pro plan looks pretty useless from now. I can see the real-time Premium requests and it's not cheap.
I'm not sure how competitve these prices are in comparison with similar services. I guess I need to figure that out on the go.

[BigBenJr]

I have the feeling that someone just cut my wings !

[jxu]

Sad if github copilot was your "wings"...

[yomajo]

What model should I set, to never worry about this limit ever (until similar announcement) again?

[JeetMajumdar2003]

GPT-4.1 and GPT-4o offer unlimited usage for the Pro Plan, while other models such as OAI, Claude, and Gemini have specific weights for premium request usage. For, using Claude-sonnet-4 (1x) allows 300 uses, while o4-mini/o3-mini (0.33x) permits 900 uses under the Plan, which includes 300 premium requests.
The Pro+, on the other hand, provides 1500 premium requests.

[knofte]

I think there's a misunderstanding somewhere. Gpt4.1 is not unlimited when it comes to premium requests, only chat.

On pro/pro+, you are still limited to 1500 premium requests even with gpt4.1.

[lucvutuan]

im using a paid account (100$/year) and when I downloaded the usage report and I saw Unlimited in Total Monthly Quota, does this mean that this feature does not apply to my account so I do not have to worry about running out of requests?

[Akash1134]

Hi @lucvutuan Could you please restart your IDE and check again? The Total Monthly Quota should now be updated with the monthly premium request allowances in effect. Thanks!

[iamcjwilliams]

Hello @lucvutuan Can you also provide a screenshot with the timestamp ?

[TMRomain]

Crazy how when you guys set that 300 hundred limits, cursor update their pricing plan from 500 hundred request to become unlimited to better align with Claude Code and other LLM providers pricing's.

[BLTCPT]

I dont like it. Now VS Code for non dev has become less effective. IF you are going to charge to use a LLM, then users should be able to bring their own.

[matt-buchanan]

I run a GitHub Enterprise instance with Copilot for Business enabled. As far as I can tell none of the usage monitoring features are available. Is this by design? This is feeling pretty half-baked at this point - what am I supposed to tell my user base? They just have to guess how many requests they have left? Or are you seriously expecting my team to run a centralised report every time someone asks what their usage is like? If you at least had an API that'd be something, but I see you couldn't even work up the energy to implement that.

In case my feedback is not clear here - this is dire. You need to try harder - right now you're just signalling that you are not investing in enterprise users.

EDIT: So I see that there is the ability to see this in the IDE - that was not appearing for us straight away this morning but after a restart or two it has started to appear. However, the lack of usage tracking in github.com is still poor, as is the lack of notice for switching this on.

[Akash1134]

Hey @matt-buchanan,

Thanks for bringing this up! Here's what's currently available for monitoring Copilot usage in your Enterprise setup:

1. Basic license reports:

   • Go to profile > Your enterprise > Settings > Billing and licensing
   • You'll see seat counts and spending by organization
   • You can download a CSV report too

   View license usage docs

2. User activity tracking:

   • At the org level: Your organizations > [Org name] > Settings > Copilot > Access
   • Shows when users last used Copilot and lets you download a report

   User activity data docs

3. Metrics API (most detailed option):

   • Provides daily usage stats by language, editor, and feature
   • Enterprise owners need to enable this in policies first
   • Data goes back 28 days

   Copilot Metrics API docs

I hear you on wanting better usage monitoring. While these tools don't give you everything you're looking for, the Metrics API probably gets you closest. I'll pass your feedback to our product team about improving individual usage tracking.

What specific metrics would be most helpful for your team?

Thanks,
Akash

[CameronRedmore]

Well, that's a shame. The unlimited nature of Copilot made it very useful.
300 credits per month, especially with some models using 10 credits per request (eg, you can use Claude 4 Opus 30 times in an entire month and that's your lot) is an unusably low amount.

I've been with Copilot since before it launched, but now, I'm looking for alternatives.

[lucvutuan]

Well, that's a shame. The unlimited nature of Copilot made it very useful. 300 credits per month, especially with some models using 10 credits per request (eg, you can use Claude 4 Opus 30 times in an entire month and that's your lot) is an unusably low amount.

I've been with Copilot since before it launched, but now, I'm looking for alternatives.

i do believe that after this update many people will think of an alternative. Especially when one other coding assistant just update their policy for unlimited for pro plan.

[ux4578]

It seems completely unfair that they would put such severe limitations on us if we were paying for it.
We would get unlimited coding and Q&A for "free" with DeepSeek, and severe limitations on "paid" Github Copilot. This is just ridiculous.

[sebhofmann]

I can see the the Quota Usage in the IDE. It currently says 1%, but i don not understand why i need to wait 30 Minutes to see how many Request i made this month. It makes it hard to get a feeling what a typical premium Request uses, compared to the monthly included Requests. I would love to have a absolute count in the IDE and also a overview for last hour, today, last week which is in real time.

I also had many Requests aborted, due some errors (mostly with Gemini 2.5 pro, which is the best model in my opinion), do these still count as premium Requests?

[Akash1134]

Hi @sebhofmann I have shared your feedback with the product team and will update you once I have an answer. Thanks!

[denysvitali]

This made my GitHub Copilot subscription completely useless. I'll not renew my subscription and switch to something else.

I'm sorry GitHub - I was a huge fan / promoter of the GitHub Copilot Pro plan and I got you a few customers too. I'll make sure all of them cancel their subscription as it's clearly not worth the money anymore.

[dreamy32]

Of course GPT 4.1 is the included model... You always have to watch it in agent mode, otherwise, it will stop. At least with Claude Sonnet 4, Gemini, etc. they act as TRUE agents that can work autonomously.

I ask him to analyse my project and make detailed documentation, and he provides me a markdown with 40 lines...

Im cancelling my Github Copilot Pro subscription to get Claude Max. 👋

[TeddyHuang-00]

Oh, what perfect timing! Just when I needed to renew my education verification—a truly delightfully streamlined process featuring their brilliantly designed auto-rejection model that clearly has my best interests at heart. So here I am, 3 days later, still waiting for this mythical "developer pack" (complete with the oh-so-generous Copilot Pro), while my "free" premium quota evaporated after a whopping 5 or 6 requests. Truly spectacular work there.

Despite being a loyal beta user since day one—because apparently that counts for absolutely nothing—I'm graciously taking my business to Cursor. But hey, I'm sure everyone will be thrilled with those incredibly generous 300 requests per month. What a bargain!

[MalleusOne34]

300 messages is a joke. I really hope it will be reverted, otherwise they will lose another sub

[Megheg694]

300 a month is just too low I am likely switching to cursor

[lucvutuan]

What plan is good for cursor? Im planning on switching to an alternative but i thought that it does not depend on the IDE but the models. So still confused.

[scottlothbrok]

The problem is, half the time the code doesn't even work and you have to correct it again. Actions that advances the code base forward, you could go through many iterations. Just to get the result you want.

This is the same with every interaction with llms like through the subscriptions of chatGpt or Claude or Gemini.
Sometimes you have to ask the models repeatedly just to do a basic task and then it will go and randomly reformat everything.

So when it comes to code and you're constantly having random stuff put in, like Claude 3.7 would randomly change button colours and layout for no reason. Fixing the errors that the models are creating, in the past you would just re prompt it to fix the error and continue, now at that rate limit. It's basically useless.

Even at 1500 requests at the Pro+ is still basically useless because you're paying for the llm to make mistakes

[Angel-William]

Documentation:

Provide examples of typical monthly usage patterns by developer type
Create a calculator to help estimate needed premium requests

[MrKnights1]

Maybe add slow premium requests after fast premium requests are depleted? 🙄

[wrek]

I would love slow premium requests as long as they're premium requests I will wait but I can't wait an entire month for my 300 to reset

[DeepDigits]

Alteast add claude 3.5 in the included models.The pro plan is useless now

[Rhystic1]

Done, like I said I would. Vote with your wallets, people. It's the only language they understand.

[smaffei]

Well…here is your feedback! Please take it seriously.

I have been using copilot with sonnet and it was really a dream. Not perfect by any means, but it was a really helpful tool. Since the change has been rolled out I have apparent run out of premium requests (I seriously doubt it, unless it was counted retroactively, in which case shame on you, but I am going on holiday now and will simply wait for the month to finish at this point).

I then gave it a go at the unlimited alternatives and I suppose I have adjust my expectations and accept an AI that consistently allucinates about the code I am working on (tells me to include files that don’t exists, explains the code making reference to functions that don’t exist, etc).

I join the chorus here: want to cap the usage? Then every request needs to give a perfect answer. Obviously that s not possible so:

• increase the limit, or adjust what counts towards the limit. Maybe if I ask to work on a single file shouldn’t count, but if the request is a complex one, with a big code as context, then ok (I don’t know, something like that)
• Please improve the gpt integration if that is your to-go AI. I am pretty sure that gpt 4 is capable of so much better than telling me “there should be a function called TheFeature() in your code that does this” when a simple search reveals no such function anywhere in the code.
• if I pass the limit I would accept a slower answer. I mean…not so long ago I didn’t even have this tool, so I don’t mind it being slower
• please include one of the claude AIs in the unlimited plan.

[jamesbalcombe83]

What a surprise the second the premium / rated mode is introduced the free models become absolutely terrible. Previously I would move between models if one didn't generate something reasonable, and GPT4.1 would be reasonable most of the time. Now its actively destructive. It ignores existing patterns in the codebase, it just dumps code into the chat, despite using agent mode, it makes nonsense mistakes repeatedly and leaves linting errors all over the place. It did not do this one week ago.

Addendum: Also all models more frequently get stuck in loops, repeating almost identical phrases over and over again making the same changes, which surprise again, eats into the premium balance.

And another thing... this happy little guy pops up again and again for no apparent reason.

[hugetiny]

My premium request has exceeded the quota, but I cannot continue to use GPT-4o as a substitute. I can only use GPT4.1.
I am a pro plan user. I think the current pro plan is totally uncompetitive.

Cursor just announced that pro plan users can make unlimited requests for the premium model. I suggest that Microsoft, as an official first-party partner of OpenAI, always provide the strongest OpenAI programming model to pro plan users without any limit. And the model selection should follow Cursor's approach. Simple tasks should be automatically assigned to the OpenAI programming model, and complex tasks should be automatically assigned to the strongest programming models such as Claude 4.

[adriablancafort]

fuk you

[jxu]

HAHAHHA vibe coder idiots! You thought they would just give you their expensive gigantic model outputs for free? Welcome to being nickel and dimed, cloud style! Maybe we'll even go back to competent programmers.

[lechibang-1512]

You're just pathetic, aren't you. Do you somehow get off on clowning people for using AI assistance during writing code?

[Megheg694]

using cursor - much better. Even their base model is useful. It works not like 4.1 in copilot, which is only good for basic things.

[jtabox]

I'm still trying to understand how this new change is in any possible way better, or at least comparable, or maybe a new and exciting thing (which is the main feeling I've been getting from the blog posts, and the various announcements in general). I can totally see myself having missed some detail that makes this whole thing make sense, so please let me know if this is the case.

Because otherwise, my customer experience just went from having a practically unlimited access to a big array of diverse options for LLMs to work with and swap as necessary, to this access now suddenly having become majorly restricted, i.e. still "unlimited", but only for pretty much the two worst LLMs from the previous array (oh and they might also be slow during prime times). And, when it comes to the rest of the models, I get 300 interactions per month, for all of them collectively, before I need to pay extra to receive the same service I've had with my subscription so far.

That's what I've understood from the whole thing, so I'm wondering what the reason was for this change. Did GitHub determine that the Pro subscription had access to more value than what it paid for? Or maybe those extra models needed to be less congested so access to them got restricted? I'm not being sarcastic here or anything similar, I'm just genuinely curious. And moreover, I can't see how this is supposed to be even remotely positive for me as the customer, let alone excited.

This feels more like GitHub either decided that $10 is too low an amount to charge for the Pro subscription, or that it could be making more money from the new models, or both. I'm not sure what to think about it, I get it that it's a company that must always make more money than yesterday. But this whole change feels a bit dishonest? Exploitative? Can't really pinpoint it, but if so, then it could also explain why it's being marketed as such a hyped, positive change for the customer.

Feels kinda bad because I've been a supporter of Copilot since quite early on, and together with GitHub they've acquired a special, familiar place in my heart (mostly in regards to other similar ecosystems, I know it's a company). All this time I haven't considered using another code assistant, and have generally been supportive of Copilot/Github in various forums and posts, but I think I'll start looking for other subscriptions for coding assistants from now on 🤷🏼

But yeah, thanks for coming to my TED talk.