GitHub Copilot is moving to usage-based billing

Hello GitHub Community,

Today we announced we are transitioning to a usage-based billing model for GitHub Copilot, effective June 1 📆 Please refer to the blog post for the full announcement link to blog, and you’ll find an FAQ below.

Important

May 12, 2026: We've updated the body of this post to include a questions about our announcement: April reports are now available to prepare for usage-based billing

📌 Frequently Asked Questions

Why is GitHub changing the Copilot billing model now?

We’re making this change now because GitHub Copilot simply is not the same product it was a year ago—it now powers far more complex, agentic workflows that consume far more compute. This change is designed to deliver a more sustainable and reliable product experience by aligning pricing to actual usage and costs.

Isn’t this just a price increase disguised as a billing change?

The per seat subscription cost of GitHub Copilot is not increasing, and code completions and Next Edit Suggestions remain unchanged. Users with intense agentic usage will likely see an increase in costs because those features consume more compute. With pooled entitlements, organizations that use Copilot Business or Copilot Enterprise may see their total charges reduced as entitlement pooling balances more active users with less active.

This just wiped GitHub’s value moat – why should I stay?

We believe GitHub Copilot remains the best value and experience for agentic coding. Usage-based billing aligns cost more closely to actual usage and value, while continuing to offer developers the freedom to choose the models and agents that work best for them.

Will there be any free models anymore with this shift?

With the shift to usage-based billing, free models are no longer part of our offering.

If we run out of AI credits mid-month, does Copilot stop working?

If your entitlement and set overage budget is fully consumed, you can continue to use Copilot for code completions and Next Edit Suggestions but will have to wait for your entitlements to reset, purchase additional usage, or upgrade your plan to use other features. We’ve built usage surfaces that show your consumption as a percentage of your budget and in dollar terms so you can track usage as you go and plan accordingly.

For Copilot Business and Copilot Enterprise users, notification at various consumption levels will be sent to your GitHub admins as your entitlements are drawn down. With pooled entitlements, all included credits are pooled across your business. This means power users can draw more when they need it, and lighter users offset the balance.

How can I forecast our AI spend if the bill changes monthly?

We’re rolling out a preview bill experience in the coming weeks to show all users how their costs may shift under the new billing model. This will be available to users via their Billing Overview page when they log in to GitHub.com.

We currently use PRUs for some features, how does this change that?

With usage-based billing, we’re replacing PRUs with a new unit called GitHub AI Credits, which are based on the tokens your interactions consume and priced according to the listed API rates per model. All features aside from code completions and Next Edit Suggestions, which remain unlimited, will be measured and billed in GitHub AI Credits.

Will Actions minutes consume my AI Credits?

Your monthly bill will include both the total cost of your token usage and Actions minutes consumption.

Additionally, Copilot code review recently moved to an agentic architecture that runs on GitHub Actions, and starting June 1, 2026, reviewing a pull request with Copilot will count against your included Actions minutes at the same per-minute rates as any other Actions workflow.

I’m on an annual plan, how does this affect my billing?

We’re retiring annual plans. If you’re currently on an annual plan, you’ll continue to use PRUs until your subscription expires. After it expires, you’ll be moved to a Copilot Free plan unless you sign up for a new monthly paid subscription.

While you can keep using your annual plan until it ends, we’ll be updating model multipliers for PRUs when our new billing model goes live on June 1. We recommend switching to a usage-based monthly plan on June 1, and we’ll provide credits that reflect the prorated amount left on your annual plan subscription.

How are you determining what to charge me and how does that change depending on which model I’m using?

Usage-based billing means you are charged based on the tokens your interactions consume, priced according to the listed API rates per model. We’ve built usage surfaces that show your consumption as a percentage of your budget and in monetary terms. These show up in your editor, on GitHub.com, and in admin dashboards.

Will you resume Copilot trials now that you’ve implemented token-based billing?

Our focus has always been on lowering the barriers to entry for anyone who wants to code and that’s why we continue to offer Copilot Free as a no-cost option. Unfortunately, we’ve observed a high volume of abuse through trials and paused them while we investigate. We are actively working on improved safeguards to prevent misuse of the trial system and will share an update when we have one.

Will we continue to see increased usage limiting?

Unfortunately, users will continue to see increased usage limiting until usage-based billing is implemented on June 1. We know it’s frustrating, and moving to usage-based billing allows us to reduce it. By aligning usage with costs, we can provide a Copilot experience that's predictable and reliable for all users.

Why not implement usage-based billing now? Why continue with painful changes like more usage limiting?

Usage-based billing requires standing up a new billing and metering infrastructure, which won’t be operational until June 1. In the meantime, we need to implement additional changes to ensure a predictable, reliable experience for all users. These are short-term safeguards, and we expect to lift restrictions once our new billing model is in effect.

Why are you only offering promotional pricing for existing Copilot Business and Copilot Enterprise users, and nothing for existing Pro and Pro+ users?

While we're not offering promotional pricing for individual paid plans, we’re offering the same 1:1 ratio of plan price to monthly entitlements as we offer for Business and Enterprise. We believe Pro and Pro+ plans remain competitive offerings that deliver high value for individual users.

Will my unused AI credits roll over from month to month?

No. Your unused AI credits do not roll over from month to month. They will reset at the start of each monthly billing cycle.

🟢 Billing Preview

Why do I have rows with 0 AIC Values?

Some code reviews entries are missing AI Credit estimations. Copilot code reviews charged directly to organizations via automation or a user without a Copilot license will reflect 0 AI Credits due to a data issue.

Based on the April usage report, my monthly Copilot bill is going to skyrocket. What can I do to control costs?

First, be sure to set up or update your budget for additional Copilot spend today. Budgets currently set for premium request spend will automatically migrate to control AI credit spend on June 1, ensuring you stay within your intended spend.

To stretch your usage, consider using lightweight models for quick questions, splitting up planning and implementation requests, and leveraging Auto Mode.

Can I use the April usage report to budget for June usage?

We recommend viewing the April usage report as a directional signal. It should give you a good indication of how you’ve been using Copilot and how much that type of usage will cost under usage-based billing.

Please see the changelog and docs for additional information.

[CodePengu1n]

Okay, with this change, will you at least bring back Opus since we're all paying for this again?

[anonimusman00-2]

The hell ? im pay it for Github Pro and Copilot Pro+ and still up the limit for Premium Requests 60$ and what i got is the f**cking limits ? are you kidding ?

This is IDENTICAL to my up-to-the-minute experience and deep annoyance. I am already looking to move to a different product or solution altogether. I started this today on an agent chat describing how confusing and nonsensical the Billing and Billing History pages are, the usage reports, budgeting -- all of it. It didn't take long for these companies to gouge the F out of devs and others.

I think they just want profit, they don't care about the server being down, they only care about money, they are very rude.

[amax1231234-sudo]

GitHub is changing Copilot to a usage-based billing model because the tool has evolved from simple code completion into more advanced AI-powered, agentic workflows that require significantly more computing resources. Under the new system, basic Copilot features like code completions and Next Edit Suggestions will remain unlimited under the existing subscription, while advanced features will be measured using GitHub AI Credits based on token usage and model type. This shift is not a direct price increase, but heavy users of advanced AI tools may see higher costs, while lighter users may experience minimal impact. Unused credits will not roll over, and annual plans are being phased out in favor of monthly usage-based billing, with dashboards introduced to help users track spending and usage in real time. For reference and web development testing purposes, you can also view related project implementation examples here: Pakwin777.

[mahomaddah]

WEEKLY LIMIT REACHED ! can not even use GPT 5 mini or Rapor
I'm paying for pro and under it it clearly written Unlimited agent mode and chats with GPT-5 mini2 also under other term it say even when weekly limit reached u can use gpt 5 mini but any way it not even have panel that show when my week will reset so I can continue. hope it will update the pricing system so I can move back to jobs

[raybreakerdusk]

Всем привет. Подскажите, как написать админу о мошенничестве, скам проект тут на сайте развернулся, я сам оказался жертвой этого проекта https://github.com/spider7zip/snoser_spiderhttps://github.com/spider7zip/snoser_spider

[raybreakerdusk]

Если что, вот моя почта topalovv925@gmail.com

[monotaskdevelopers]

At least bring back Opus 4.6? 4.5?

[RichMercer]

They're priced at the same as Opus 4.7 though, so there's no point.

[M4XEURS]

They're priced at the same as Opus 4.7 though, so there's no point.

So, can the Pro version still use 4.6?

[imm]

They're priced at the same as Opus 4.7 though, so there's no point.

People online say that Opus 4.7 uses more tokens per request on average than 4.6, so it's not quite obvious what is better.

[ktalanov]

They're priced at the same as Opus 4.7 though, so there's no point.

So, can the Pro version still use 4.6?

Nyooooope.

From the article that we are all discussing:

3. Opus models are no longer available in Pro plans. Opus 4.7 remains available in Pro+ plans. As we announced in our changelog, Opus 4.5 and Opus 4.6 will be removed from Pro+.

Huh, that's weird. I feel like I would have noticed that. Okay, let me click it.

Ah, okay. If you have Changelog emails turned on, or follow their Blog, you may have noticed a changelog entry titled Claude 4.7 is Generally Available" and, like me assumed "okay, they released 4.7 for Copilot, I'll check the multipliers later."

Wait, but they said they informed us that Pro users would lose access? I don't see anything like that after skimming through the changelog. Well, if you actually read the entire link at a snail's speed, you can sort of cobble together what Microsoft PR tried to do 11 days ago.

It seems like they are not going to be providing any Opus models (the biggest visible announcement in that changelog was that Opus 4.7 would be provided at a promotional multiplier at 7.5x for a few weeks, making several users assume that it was a typo -- surely it's not correct that the regular multiplier rate will be 22.5x?) Well, it's worse, it's actually 27x.

To go back to @M4XEURS question, a few days after they posted the announcement about Opus 4.7, all Opus models were silently removed from Copilot. We're on version 1.0.39 and I remember 1.0.36 already didn't have them.

So where was this notice they talk about? Well, I've read the changelog three times to make sure I didn't miss anything. It appears that what they mean by "we announced in our changelog" is two sentences, one of which is buried in the second paragraph and the second appearing after a subtitle and a video. If you read them together, you might divine the meaning.

It's a sneaky attempt, but Microsoft still has a lot to learn from BP and Shell. I can see their thinking though, this is exactly where someone looking for answers as to why Opus suddenly disappeared from their paid Copilot version would look first, in their "Opus 4.7 is now available!" email.

@M4XEURS these are the sentences they expected (edit: they obviously didn't and just buried the announcement) you to add together and after a few minutes' of mental arithmetic come up with the conclusion that all Opus models have been removed from Copilot Pro.

As part of our efforts to improve service reliability, we are streamlining our model offerings. Over the coming weeks, Opus 4.7 will replace Opus 4.5 and Opus 4.6 in the model picker for Copilot Pro+. We’ve seen strong improvements across our benchmarks, and we’re committed to providing individual users with state-of-the-art models while ensuring a fast, reliable Copilot experience

If you scroll past all the crap following the paragraph, you might notice:

Claude Opus 4.7 will be available to Copilot Pro+, Business, and Enterprise users.

And if you combine the two sentences together in your mind, then drop the "assume good faith" thing Wikipedia has been teaching us, you might come to the conclusion that Opus models are going away from Copilot Pro plans. Ah, yes. Clear as mud.

I don't know about all you dullards, but I know that if my software is losing a major feature, there is nowhere better to announce it than by sandwiching it into a a "changelog" about a NEW feature. Can't wait for the email from my toaster manufacturer with the subject "Congrats! Your toaster now consumes 45% less electricity!"

Your toaster now tells you the weather for the upcoming day when the handle is pressed and consumes 45% less electricity! This firmware upgrade will be pushed through automatically when toaster is next connected to WiFi. As part of the upgrade, the bagel/poptart/bread toasting functions will be consolidated into NeoToast, which will be available to anyone on a ToasterPro+ plan or above.

This was your notice:

I have never seen a company be so blatantly disingenous with their customers. Forget evil, this is less malice and more incompetence. It's like the VP of Communications found a copy of Catch-22 on the table at Starbucks and read it as an instruction manual. "Oh, you have an annual subscription? We're not raising prices, we're simply telling you that you now have 33 Sonnet 4.6 premium credits instead of 300! We didn't remove your access to 4.6, we simply "consolidated" it into Opus 4.7 (note; you will not have access to the "streamlined" Opus models, even if you previously did. Where did we announce that? Well, in the changelog that implied you'd get access to Opus 4.7, obviously.)

You know, I understand companies can be evil, or rather that I should not expect them to act in by best interests. I even understand that at $8.33/mo on an annual Pro plan, you will need to change something so you're not losing money now that LLMs are expected to be agentic, call tools, have a larger context windows and take longer autonomous actions on one request.

I got my annual subscription in August of last year. By December, I realised they would not be offering a $10/mo subscription if one premium credit could mean anything from "List the first five prime numbers" to "Why Darmok and Jalad at Kalenda's? Why not Darmok and Kalenda at Jalad's?" to "Create an MVP of Amazon for dogs, including all the specs of a physical product like an Amazon Dot, so that any time the doggo wants treats, they can just boop the "treat button" that instantly dropships dog streats. Finally, when MVP is done, send an email to the top VCs in my area. Make no mistakes."

I have three months in my annnual plan. I just figured that they'd let me ride out the rest of my few months -- what would it cost to them to have three months at the old cost, and simply give the new cost to new/monthly customers? Obviously for the tail end of the distribution quite a lot, but consider the PR and reputation loss from this cluster.. vs. the number of companies who would actually trust Microsoft more if they honoured their commitments. They could still switch most people, who are on monthly plans!

(btw, @microsoft, we actually have something in common: I also am really bad and marketing and not knowing what my users want. Can I get hired? From what I've seen over the last 48 hr I wil fit into your company culture perfectly and I'm pretty sure you can just give me production access on the first day. Don't worry, I know the Status/Outages page must stay super-hidden, I got you.)

Anyway, for anyone on a Pro plan this is your notice that your Opus models went away a few weeks ago. Sorry, I meant were consolidated into a model you don't have access to.

You know, if you any of you fall on hard times you practice that in the mirror next time you consider robbing a house:

I swear officer, I was just consolidating their newest belongings into the back of my truck!

I don't really have anything further to say. If you've read this far, congrats, I guess? And may god have mercy on my soul for making you do so.

[oevillegas]

Github to Pro users: Sorry sir/madam, this is only for paying customers!

[joel-surfthing]

Can I request a refund for the remainder of my annual plan? I've stopped using Microsoft GitHub Copilot altogether.

[Shah-Nisarg]

is a refund available to google play subscribers?

Hi @Shah-Nisarg, yes a refurn is available for Google Play subscribers, but I believe it needs to be processed via GitHub Support and is not self-serve. We're looking into this and will comment here again with specific guidance

[yht1024]

How can an annual user convert to a monthly subscription? I haven't seen a clear way yet. If you want to cancel and renew your monthly subscription before May 20th, but it seems that once you cancel, you won't be able to subscribe anymore, or should you wait for a channel to convert after June 1st

[WHBSmith]

How can an annual user convert to a monthly subscription? I haven't seen a clear way yet. If you want to cancel and renew your monthly subscription before May 20th, but it seems that once you cancel, you won't be able to subscribe anymore, or should you wait for a channel to convert after June 1st

I had a dig around for this too it's in Billing & Licensing -> Licensing and then on the right you'll have your subscription and a 'Manage subscription' drop down. If you click on that you can switch to monthly billing or request a refund. Probably wait until June to do any changes or I guess you won't get any pro-rata credits.

[snakex64]

Literally why they're increasing the annual plan's premium usage request. In hope that they can screw people over hard enough that they cancel the annual plan so they stop loosing money on the old billing model.

Disgusting practices

[thmsbernard]

Is there a benefit of using the Copilot Pro+ at 39$/month instead of using the Copilot Pro at 10$/month and paying for extra usage?

[iayoung85]

Yeah it's tough to see what the $10 gets you as the credits don't rollover vs. PAYG with Claude etc. Maybe it's basically just convenience with different model access?

yeah as far as i can tell reading into all the posts.. all you get for pre-paying $39 is the privilege of talking to opus or GPT 5.5.. No real value adder.. I agree that the existing requests system was broken.. but I think this is essentially losing 90% of the value. For me the ability to choose different models was cute.. but not helpful.. often just increasing confusion.

[ManojSurangaBandara]

For those dont understand what's happenning, Github is charging you 39$ to start using high tier models and 10$ to start using rest. If you surpass that $ amount you pay the extra, if you didn't its all gone....simple !

[Slix-M-Lestragg]

Seems like Opus model will be only for Pro+ plan https://github.blog/news-insights/company-news/changes-to-github-copilot-individual-plans/?utm_medium=email&utm_source=github&utm_campaign=FY26APR-WW-LCM-BLA-ProPP-IND-Dev-TX-USGCHGMTH
Opus models are no longer available in Pro plans. Opus 4.7 remains available in Pro+ plans. As we announced in our changelog, Opus 4.5 and Opus 4.6 will be removed from Pro+.

I think that is stale, given on the new pricing you're paying for usage anyway, there doesn't seem to be a reason to hide high end models behind a higher tier which you're getting equivalent credits for anyway.

I fully agree, seeing the move they did by replacing opus 4.5 and 4.6 with 4.7 only and at 7.5x, I'm not surprised

Guess I should have said 15x ...

[Daniel15]

Higher credit at the start of every billing cycle? Lower multiplier for each request, perhaps? I have no idea, either. It could also be that paying for extra usage is charged at a higher rate than the monthly billing rate.
That's all speculation, of course, so I hope GitHub clarifies this for us.
Reading some of the screaming in other threads about other changes GitHub is making, I wonder if this is going to cause more harm than good.

$10 for $10 worth of usage at API rates is simply unusable unless/until API rates come down. There's zero reason to use GHCP after June 1st if this structure goes through as planned.

@OotTheMonk Everything is going to go this way at some point. Subscriptions are heavily subsidized at the moment, which is a major reason why neither OpenAI nor Anthropic are profitable. It works while they're growing and investors are throwing money at them, but it isn't sustainable. The API pricing is closer to what it actually costs them to provide the service, but as far as I know even the API prices aren't profitable at the moment.

I think Github Copilot moving towards being more like a PAYG plan is probably the beginning of an industry-wide movement over the next year or two.

[KnowlesLabs]

Nobody wanted the api token pay model from the start and nobody wants it now. It was nice everyone got used to tools they won't be using in the future. Microsoft should never have been the pick to head this, slice it anyway you want it blows. stop giving the same 5 people all your money. I never did trust the anthropic, openai, google we ain't together but we together orgy.

[science64]

how much would be the cost of the tokens? now premium request is 0.04 USD, is it going to be same or 1 million token will be around 10-20 USD and will we have any information how much token agents are using?

[ShepAlderson]

I expect that there will be granularity down to the penny for billing. So, let’s say Pro at $10/mo gets 1000 AI credits. You ask a question that uses $3 of API billing, so it charges down your AI credits by 300, so you’re left with 700.

For some context, a single session with something like Opus or GPT-5.5 while doing a plan and implement can easily run into the realm of $30-40. For context, I did a plan of a pretty substantial feature today using API billing and Opus and it cost about $35 for that single chat session. (This was in Claude Code, not Copilot.)

With the pricing change, unless you use a ton of autocomplete suggestion stuff in your editor and that’s your main surface for interacting with Copilot, or if you have no other option at your work but to use Copilot, you’re better off with almost any other plan. If you like GPT models, that means going to OpenAI directly, or if you like Claude models, Anthropic.

I’m sad this change is happening, but I am not surprised in the slightest. I can only imagine that agentic workflows have GitHub hemorrhaging money, and while Microsoft did invest a lot in OpenAI, they are still apparently dealing with API pricing for the models GitHub serves? I think the only way a company can be competitive and offer a substantial value to compete against “Just to pay for the API yourself” is to train and serve their own models. We’ve seen that happen with Cursor, and I expect we’ll see more. (I thought that was the intention of Raptor Mini, but who knows what happens with that now.)

Also, the fact that the “AI Credits” expire at the end of the month means you’re almost certainly better off just going to the API providers directly and just paying there. If you use less than $10 or $39 in inference with your AI Credits in a month, you’d bet better off just paying à la carte API prices.

Now is a good time to start experimenting with agentic harnesses that aren’t connected to a specific provider’s service. (Things like opencode, for example.) Luckily the open weight models are getting better and better with each release, and hopefully we’ll continue to see reasonably priced API access for those continue.

[eric-rolph]

I installed Claude Code and Codex VS Code extensions. Much better experience consumer wise.

[ceelian]

But if someone is already invested in the vscode/github copilot ecosystem, they could always use bring-your-own-model-key (BYOK) and have API pricing of the model provider directly. So even if it is a price increase, it is still quite flexible.

[snakex64]

But if someone is already invested in the vscode/github copilot ecosystem, they could always use bring-your-own-model-key (BYOK) and have API pricing of the model provider directly. So even if it is a price increase, it is still quite flexible.

You have the flexibility of paying more in every single other scenario, congrats

[PanAchy]

@science64 I made github.com/PanAchy/copilot-arewecooked as a way to get a sense of what your cost will be starting June based on your historical usage.

[Omar5962]

tengo una suscripción Copilot Pro o Pro+ y organizacionalmente cuento con todos los servicios Microsoft Copilot 365, como aplica esto para mi.

Machine translation:

I have a Copilot Pro or Pro+ subscription and organizationally I have all the Microsoft Copilot 365 services, how does this apply to me.

@Omar5962, these changes aply to GitHub Copilot and not to Microsoft Copilot 365 services.

Machine translation:

estos cambios se aplican a GitHub Copilot y no a los servicios de Microsoft Copilot 365.

[sudo-js-py-c]

u forgot an p there in reality it is "apply"

[scbritton]

I'm a bit confused by this. I currently have CoPilot Pro which gets me a set number of Premium Requests. Under the new system, that gives me US$39 a month in AI Credits.

How is this different? Is this a shift to some kind of time-based system where requests that take longer cost more? What does this mean in plain language?

[MarkBellaCoreCEO]

Wow, what a bunch of greedy Fuc*ers

[MarkBellaCoreCEO]

Good Ole Claude just recommended this for me lol.
My Honest Recommendation
Install Roo Code (free VS Code extension, search "Roo Code" or "Cline" in extensions)
Configure it to hit LM Studio at localhost:1234 — I'll write the config file for you! Free for 95% of sessions
For sessions that need Claude-level intelligence (hard bugs, architecture decisions): load OpenRouter credits, ~$20 tops for a month of heavy use Keep Continue for autocomplete while you type Total monthly cost: $0 to $20 depending on how much you lean on cloud models.

Roo Code (VS Code extension — already in your VS Code marketplace)
Closest match to what you do now. Full agent mode.

Reads files, writes files, runs terminal, chains tasks — identical to Copilot agent
Backend is your choice: LM Studio local (free), OpenRouter, Anthropic direct
Point it at http://localhost:1234/v1 + whatever model LM Studio has loaded → $0/month
If local model isn't smart enough for hard tasks: add OpenRouter credits ($20-40/month actual usage, pay-as-you-go, not subscription)
OpenRouter has Claude Sonnet 4.5 at ~$3/million tokens — a heavy session costs $0.50, not $50
This is your answer. It's Copilot agent but you control the backend.

Cursor ($20/month flat)
VS Code fork — separate app, not an extension
Drag your workspace folder in, imports your extensions
Has full agent mode, multi-file edits, "composer" for autonomous tasks
$20/month hard cap, no overages
Uses their own Claude/GPT-4o access in the subscription
Downside: different app, small migration friction, still $20/month forever

Windsurf ($15/month flat)
VS Code fork by Codeium, separate app
"Cascade" mode is their agent — legitimately good, autonomous multi-file
$15/month hard cap
Better than Cursor for pure autonomous coding tasks in my read
Same downside: separate app, still a subscription

[azimaro]

Mark, we should trust your free gift aftyer you called us greedy fuc*ers? lol (jk)

[GiuseppePiscopoHypex]

Regarding Roo Code (which I haven't used yet): just watch out for a (short?) potential rough time while the previous teams hands over to a new community team (link).

[seancondev]

Good Ole Claude just recommended this for me lol. My Honest Recommendation Install Roo Code (free VS Code extension, search "Roo Code" or "Cline" in extensions) Configure it to hit LM Studio at localhost:1234 — I'll write the config file for you! Free for 95% of sessions For sessions that need Claude-level intelligence (hard bugs, architecture decisions): load OpenRouter credits, ~$20 tops for a month of heavy use Keep Continue for autocomplete while you type Total monthly cost: $0 to $20 depending on how much you lean on cloud models.

Roo Code (VS Code extension — already in your VS Code marketplace) Closest match to what you do now. Full agent mode.

Reads files, writes files, runs terminal, chains tasks — identical to Copilot agent Backend is your choice: LM Studio local (free), OpenRouter, Anthropic direct Point it at http://localhost:1234/v1 + whatever model LM Studio has loaded → $0/month If local model isn't smart enough for hard tasks: add OpenRouter credits ($20-40/month actual usage, pay-as-you-go, not subscription) OpenRouter has Claude Sonnet 4.5 at ~$3/million tokens — a heavy session costs $0.50, not $50 This is your answer. It's Copilot agent but you control the backend.

Cursor ($20/month flat) VS Code fork — separate app, not an extension Drag your workspace folder in, imports your extensions Has full agent mode, multi-file edits, "composer" for autonomous tasks $20/month hard cap, no overages Uses their own Claude/GPT-4o access in the subscription Downside: different app, small migration friction, still $20/month forever

Windsurf ($15/month flat) VS Code fork by Codeium, separate app "Cascade" mode is their agent — legitimately good, autonomous multi-file $15/month hard cap Better than Cursor for pure autonomous coding tasks in my read Same downside: separate app, still a subscription

I'm using cline - its good - you just have to be very organised, and set rules up. but with the addition of deepseek v4-pro you can spend ~$4-5 developing out a complex feature with a claude like model, then 80c getting deep seek to implement - If you are very brave you can ask deepseek to do it all for $1.

I just implemented ~4000 lines this way, overall its okay; needs a bit of tidyup but it does work.

[fastupmedia]

I have a pro+ plan, I used 4.6 opus before it got removed, now trying out gpt codex 5.3. If I use let's say 90% current premium request during a month, how much would this cost me extra? Anyone know any estimates or will premium request still be included so I still get my 1500 request ?

[rayrite]

Opus 4.6 will soon cost 27x after June 1. So you'll get roughly about 140 or so requests a month of opus if my math is correct (3900 AI credits/mo. ÷ 27)?

[fastupmedia]

But if we pay as we go, with a large project will that not cost shit load of $$. I can use gpt 5.3 codex aswell but worried that pay as you go will cost thousands of $ vs 1500 premium request

[lmf-git]

But if we pay as we go, with a large project will that not cost shit load of $$. I can use gpt 5.3 codex aswell but worried that pay as you go will cost thousands of $ vs 1500 premium request

Yes.

[ShepAlderson]

But if we pay as we go, with a large project will that not cost shit load of $$. I can use gpt 5.3 codex aswell but worried that pay as you go will cost thousands of $ vs 1500 premium request

Yeah… I think folks are going to be in for a shock when the “usage cost preview” lands in May. If you’re doing agentic coding, where you set an agent on a task with a checklist or plan and then let it run for 10-15 minutes, something like 5.3-codex can easily burn through $10-15 in API credit costs, or if using 5.5 or Opus, think $30-40. Literally a single prompt could eat your entire allotment of “AI Credits”. I’m curious to see how they will handle it if a task isn’t complete. Will they do like Anthropic and stop the agent cold, just failing halfway through, or will they allow the agent to at least finish?

[rglidden]

Will there be changes to allow a single GitHub account to have both an Individual Pro/Pro+ subscription as well as a seat assigned by one or more Businesses or Enterprises? Or do individuals still need to create a separate GitHub account for each to prevent the wrong Business being charged, or being charged for personal use?

Hi there 👋

Copilot Pro or Pro+ cannot currently be used on the same GitHub account that also has a Copilot Business or Copilot Enterprise seat assigned.

For now we recommend using separate GitHub accounts to keep personal and work usage clearly separated and to help avoid any unintended billing confusion.

[rglidden]

That's not what the documentation recommends.

GitHub recommends that you use one personal account for all your work on GitHub.com. With a single personal account, you can contribute to a combination of personal, open source, or professional projects using one identity. Other people can invite the account to contribute to both individual repositories and repositories owned by an organization, and the account can be a member of multiple organizations or enterprises.

The whole way Copilot is billed and managed goes counter to that recommendation.

[lipemat]

Will unused AI Credits roll over into future months or are the first $10 or $39 in AI Credits a "use it or loose it" scenario?

[Tpeterson37]

The fact they openly just agreed that "Yes, it's usage-based but.. if you don't use it, we're going to take your money anyway" is such a POS move.

[liuyxpp]

Hi there,

It should have been mentioned in the FAQ right away.

Valid point, we've added this question into the discussion announcement.

Your unused AI credits will not roll over from month to month. They will reset at the start of each monthly billing cycle.

It is like you expect someone with brain-defect to accept this deal. Bullshit!

[iroach]

@nardobarbosa Are unused AI credits at least applied to other users usage? The answer to that question should be added to the FAQ.

[GiuseppePiscopoHypex]

@nardobarbosa Are unused AI credits at least applied to other users usage? The answer to that question should be added to the FAQ.

The linked blog post speaks about "pooling" in the context of business plans, not individual plans. The way I get that is - during the month - there's a pool of credits that can be used by many users, whether high-usage ones or low-usage ones. Not sure if that's what you're referring to.

[GiuseppePiscopoHypex]

believe me 10 USD credit, you may even finish in a day and with one request. this is truelly so bad.

I was also thinking that it has been mentioned that a new billing tool will be available at the beginning of May. That should show your hypothetical spending in past months, would the usage-based billing already have been in place.

[Domush]

Removing the free models destroys any value in the individual plans

[cyberjob001]

I might just unsubscribe because of this very reason! Copilot was the only sensible package provider, but seems they want to follow in others footsteps! Maybe the change might be better, but removing free models. I am out!
Why pay for multiple different subscriptions, if only advantage present is removed.

[disi]

I was always saying that the problem with other AI is that in the middle of a project you run out of credits and don’t even get code completion.
I have the free tier at work and can say that coding assistant still works if you run out of credits. But no chat.
Will it be similar with my personal pro account? Just more credits?

[omar-alsaheb]

I might just unsubscribe because of this very reason! Copilot was the only sensible package provider, but seems they want to follow in others footsteps! Maybe the change might be better, but removing free models. I am out! Why pay for multiple different subscriptions, if only advantage present is removed.

So what are you using today?

[cyberjob001]

I was relying on gpt-5 mini after the requests were depleted. I already have codex and claude subscriptions. Only kept copilot because of free models. Now I wont need copilot and will move to some other provider which provides fixed number of calls rather than token based usage.
If anyone even provides a basic model which can be used as fallback when premium requests are depleted i would lean towards it.

[movy]

You can enable budgets on your account and keep working even after your premium requests have depleted (until you hit budget quota you set for yourself). This is how I figured that Pro+ will be more cost efficient for me based on budget I burned beyond normal Pro subscription request allowance.
https://github.com/settings/billing/budgets?utm_source=vscode

This is actually GREAT experience and I always felt sorry for users of other coding agents, who got overcharged beyond what they planned to spend on a single beefy prompt.

[AvinoamSebbah]

What are you doing to do for GitHub student account ? About GitHub Ai Credits
Thanks

[Himpq]

They have suspended the registration of Education Benefits, and I also want to know the next steps

[shyeyian]

• Request-based billing is user oriented.

  • Users do not care about how the AI works, they only care about the input and the output.
    • (Just like users do not care about the implementation of an function, but only care about the declaration int func(int value)).

• Token-based billing is not user oriented.

  • It exposes internal parameters (which is so-called 'tokens') to users.
  • This is a bad design in software engineering, and will be especially bad in GitHub.

Please, move back into request-based billing. Thanks.

[lusp75]

Request based billing is awesome because if you do it correctly you can get Claude to write an entire application in a single request.

Clearly that isn't sustainable for MS, who is paying the bills for this functionality. I wondered how long it would last, bummer it's over. Github was the cheapest tokens by far.

Actually, there was Trae, which was cheaper (600 monthly requests for $7). The problem is that even those crafty lot have switched to a token-based system. Damn them

[DreamFulFil]

It's impossible for them to move back. They used "Request-based billing" to bait you to subscribe and moved to "Token-based billing" with all the excuses their AI comes up with.

[brandon-fryslie]

Request based billing basically locks in a system where either a company is either charging way too much for tokens, or way too little. That's not a workable business model. Whether that's an excuse or a reason depends on your point of view. I think it's just a fact.

[seancondev]

Request based billing is awesome because if you do it correctly you can get Claude to write an entire application in a single request.

Clearly that isn't sustainable for MS, who is paying the bills for this functionality. I wondered how long it would last, bummer it's over. Github was the cheapest tokens by far.

this is true - but I found that with the extended sessions even the best models lost coherence and started developing on hallucinations - I ended up stopping them and reframing. so I'm not sure how useful is actually was

[jch254]

Guess this is firing up https://github.com/juliusbrussee/caveman

[matteocacciola]

I am really sorry but I am not understanding very well the "convoluted language" you are using. I have simple questions:

1. what will happen if I will not do anything with my individual Copilot Pro plan?
2. why should I move to a monthly based plan before the expiration of my subscription?
3. why don't you provide a yearly based subscription, even if on an usage-based policy?

[M4XEURS]

You will face very very very high multipliers like 27x, and you will not get your opus back.
But if you move to monthly, opus is also not for you, only for pro plus.
lol

[matteocacciola]

bye bye Github Copilot Pro

If you're currently on a monthly subscription, you will be automatically be moved to usage based billing on June 1st.

[sudo-js-py-c]

Why isnt there a Middle Finger Emoji? "Github Community Admin"
It would be a great feature i believe

[ariajanke]

"AI credits"?! More number games. Is there just no where for me to have anything approximating a professional coding environment now? I'm done!

[ariajanke]

I hate github.

[cyberjob001]

What happens next month, are they rolling up the unused credits?
They are just throwing us away!

[mitchelsellers]

I finally got access to the report earlier this morning; however, with access to the report, there are more and more questions now.

Traceability & Identification

The report is "good" it tells me how far off I'm going to be next month. However, I'm still lacking the most critical answer that I need.

What was the cost of this action that I asked Copilot to Do

I have no visibility into the actual tokens used for an individual interaction. Sure, my usage was high, but I've sent some crazy stuff to Copilot and more reasonable, which of those really had the spend? What was the spend. Some of these things, are 100% worth throwing real money at. But others might not be.

My personal spending on Copilot Code Review for example charted out at more than $150/month, which seems a bit high, but I have a lot of volume.

Enterprise vs. Individual

Today's blog post outlines a flex amount for individual users.

The docs for Enterprise/Business customers discuss a temporary 3 month granting of additional credits, that seem to mimic the Flex amounts, but they are only truly limited time.

This brings two questions:

1. Why are we penalizing corporate subscribers by only giving 3 months, while giving this to individuals. This is especially h ard for small teams (< 10 developers for example)
2. Or is this a sign that the individual developers should be ready for the hammer to drop in 3 months on the flex

I understand that change needs to happen, but PLEASE give better visibility into what things will and ARE costing BEFORE you do this. The knee-jerk reaction of disabling and stopping usage of features will be far greater without simple transparency.

[Captain-RedArcher]

Amazing...
I was blocked with some bull&&&& weekly limit, which is in the terms.
I had at least 8% left on premium usage and was not just blocked from premium but from all usage.
So the 10th was my renewal date, but I already canceled due to the new policy coming in June.

Now that I'm going to use my last 8%, I lost the premium credit that I paid for. I only have basic usage now.

Just like Microsoft. Typical Hen House Horse&&&&.

Instead of giving a reasonable product at a reasonable price, I went and paid extra over at Claude and have access to Opus 4.7.
Two days using it and all the bugs in my emulation have been fixed; GPT 5.2 and the watered-down Sonnet 4.6 went around in circles, which cost me 92% of premium credits.
By design, no doubt to nickel and dime Copilot customers.

[chiheiro]

1 AIC = $0.01

I even started using it from April 14th

[Captain-RedArcher]

Here is the reality.
You will be charged over 3X the amount for the same runaround as you get now.

Charge you more for the same old broken models.

[paul-ennemoser-jaha]

I can't download the CSV... it errors out:

[KnowlesLabs]

Did you try upgrading your plan? The feature may only be error free if you pay enough.

[llz0717]

Just try again, too many people are downloading their report now

[GlenBoonzaier]

So, if I’m understanding this correctly, the logic is basically:

“AI has finally become genuinely useful and integrated into people’s workflows, so now it’s time to monetize every possible angle. We’ll charge users for the tokens they use, profit from the tokens they don’t use, and hide it all behind ‘usage based billing’ so it sounds reasonable.

The reality is that this model creates a system where the provider wins no matter what. If users are cautious, you profit from unused allocation. If users fully embrace the product, you profit from overages and escalating usage costs. Either way, the customer carries all the downside.

Calling this ‘fair usage’ or ‘flexible pricing’ feels disingenuous when the real outcome is predictable: people become anxious about using the tool freely, despite already paying for access in the first place.

GitHub Copilot was supposed to improve developer productivity, not turn every interaction into a meter running in the background.”

[ep-hansenm]

It would be very helpful to have more guidance (and maybe some examples) on how to configure pooling for an enterprise with multiple cost centers.

Example

Let's say I have an enterprise of 100 users (with Copilot Business seats) separated into three cost centers:

• cost center A has 20 users and is not willing to pay more than the 19$ per user per months
• cost center B has 30 users and has a budget that allows to pay at most 40 $ per user per month (19$ base price + 21$ additional credits)
• cost center C has 50 users and has a budget that allows to pay at most 30 $ per user per month (19$ base price + 11$ additional credits)

How would one configure this so that...

1. ...none of the cost centers gets billed more than their actual usage
2. ...each month, each of them is guaranteed to be able to use the AI credits included in the base price that the cost center paid for (e.g. cost center A must be able to use a total of 380$ (20 * 19$) of credits)

?

[QSQ-QAQ]

如果是这个账单，我想我应该会选择使用其它产品，这个账单对我来说这简直就像天塌下来了一样！

[dalenicholson]

I was able to get my billing preview, but can I get that for prior months too? What about for this month so far? Last month was a really light month for me, so would appreciate more data. It showed I would have paid about $105 last month.

edit: also, I forgot to ask, I see we can set budgets but it's not at all clear to me how to do that now for next month. It looks like I would be setting up for overage if I ran out of premium requests still. Does that mean if I set a $200 budget that at the end of this month it would automatically roll over to the new plan methodology?

[QSQ-QAQ]

At present, they only have the bill for April.

[srinathganesh1]

GPT 4.1 was a free model with 0x usage multiplier. So what happens to that? As per pricing table https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing#pricing-tables looks like it will NOT be free anymore?

[AceVentura]

No more free models in the token based billing

[hrkisk]

wtf Github?

[aidannairn]

59x w/ no overage… Wow, I’m genuinely impressed! 🤯

[userofit123]

where is everyone getting this preview from? All I found was the place to download the CSV from. Where is this interface?

[aidannairn]

where is everyone getting this preview from? All I found was the place to download the CSV from. Where is this interface?

https://copilot-billing-preview.github.com/

[shaunbruno]

As others have pointed out, this isn't the win for developers it's being framed as. The heart emojis are coming from people who either benefit from the change or haven't thought through what it actually means.

Before: predictable cost

You picked a model, you knew the ceiling. Good Model X cost 12 cents whether the request was simple, complex, botched, or timed out on the server side. Model performance was still a slot machine — we accept that — but the cost of pulling the lever was knowable before you pulled it.

Now: the vendor decides what you used

Under usage-based pricing, the corporation decides the actual usage. You pick Good Model X at 3x the base rate — 3 cents, seems fine — submit your request, and it comes back at $1.

What happened in between is opaque:

• The model decided it needed to scan more of your repo. Input tokens climbed.
• The model decided to produce verbose output. Output tokens climbed.
• The system decided the scope, decided whether to complete or time out, and decided your bill.

Your only options after the fact are accept it and keep using the service, or accept it and stop.

The 1-cent credit is sleight of hand

Pegging an AI credit at a penny makes everything sound reasonable. It isn't — because you have no control over the upper bound. The monthly budget cap is necessary but not a real solution: "John can't use AI for the rest of the month because one misinterpreted request ate his credits" is not a feature. And pooling credits across a team doesn't fix it — it socializes the unpredictability. Now John's bad request is also Jane's missing credits.

The car wash test

At a car wash, you know the Ultimate package costs 3x the Basic. You don't know if it'll actually clean your car better — that's the gamble. But you do know what you're paying when you pull in.

Imagine driving out and finding a charge for 10x the sticker price, with no way to have known that was possible. That's the model being celebrated here.

Bottom line

The only guaranteed winner is the vendor. Anyone telling you otherwise is selling something.

[kctherootkey]

Thank you, by looking into the predictable cost of what I will pay in the future I will immediately cancel the GitHub Copilot subscription end of May and switch fully to Claude Code, that's for sure!!

[rglidden]

With pooled entitlements, organizations that use Copilot Business or Copilot Enterprise may see their total charges reduced as entitlement pooling balances more active users with less active.

😆

I was expecting an increase. But I did not expect the overage/usage portion to be over 22x higher when 75% of our users are currently staying below the included 300 PRU's under the current model.

[rglidden]

And that's including the "promotional amounts". Which means in 3 months, that cost would jump by another $2700 as the included usage drops from 3000 AIC to 1900 AIC per user.

[ktalanov]

@rglidden 's results are impressive, here are some results from a lonely $10/mo Pro subscriber so we can at least cover both extremes of the spectrum.

Based on my own results, it appears they were bleeding money from Anthropic specifically. I might create a little browser extension that works completely locally and gives you text output (something like below) so we could compare how much we'll be getting screwed. (By code, I obviously mean get the last good use of my PRUs and vibe code it. Don't worry, I'll tell it to make sure not to make any mistakes.)

For myself, the data belw is for the entirety of the month of April. Several things stand out to me.

First, I've used 43.66/300 PRUs in Aprll, for a total of $1.51 (if charged at the overage rate of $0.04/PRU). Under the new system, I would be charged $55.69 (I refuse to add the 3-month discount as I feel it skews the results to make them seem lower even though in September you'll be paying more so I don't want our comparisons to appear lower than what they'll actually be.)

Next, let's talk about the difference in model pricing (non-multiplier wise, just in terms of AIC):

Average AICs per request for Claude Haiku 4.5:  44.332  Average Cost per request:       $0.44 (0.66 PRUs, i.e. 2)
Average AICs per request for Claude Opus 4.6:   398.869 Average Cost per request:       $3.99 (6 PRUs)
Average AICs per request for Claude Sonnet 4.5: 55.029  Average Cost per request:       $0.55 (**1 PRU**)
Average AICs per request for Claude Sonnet 4.6: 84.298  Average Cost per request:       $0.84 (27 PRUs)
Average AICs per request for GPT-5 mini:        0.063   Average Cost per request:       $0.00 (6 PRUs)

Total PRUs consumed:                    43.66 PRUs
Total PRU cost (@ $0.04/per):           $1.59 (not billed, just shown)
Total AICs consumed:                    5,568.90
Total AICs cost (@ $0.01/per):          $55.69

The multiplier for Opus 4.6 in April was 3x (this still makes me laugh, or weep, or something -- I ended up not using it because I felt it was too expensive and I didn't want to go over, using mostly Sonnet 4.6 and ignoring the frontier model.. and ending the month having used only 15% of my PRU allowance. Anyway, the Opus line means that only TWO requests to Opus 4.6 cost $4. Assuming they're actually being honest (plus or minus the house's edge) in terms of their biling, that meant they would be paying $1,200 for a Pro+ user with 2,000 PRUs or $200 for a Pro user with 300 PRUs (based on $2/PRU and 100 PRUs if you only used Opus).

it looks like Sonnet 4.6 was bleeding them dry also, at 1x multiplier it was costing almost as much as Opus.

P.S. I wish this announcement came just one or two months earlier (I understand why it didn't, but they could have at least given us a heads up and just set huge multipliers for Anthropic models and given us 12 weeks of notice instead of 5. Next, holy crap. Did I miss the boat. Looking at my "billing", for most of the last few months I've used about 10-20% of my PRUs. I figured I could get through the Seasonal Affective Disorder and come May, be in a more productive mood. Then.. comes May. Worse yet, if I actually had the energy (and remembered that it was a Microsoft product), I could have simply added up the tokens on a single-PRU, did the math and switched to the Pro+ plan and worked non-stop using Opus for the short time we had it and Sonnet for when we didn't. Even then, I would have thought they'd at least give me the three extra months considering my annual subscription ends in August. I remember how everyone knew exactly the tricks Microsoft pulled (being a big Linux nerd and excited about a new OS to geek out at at 14, I remember the Halloween memos, "Micro$oft", and "embrace, extend, extinguish". I guess not much changed, eh?

Edit: I can't type

[oevillegas]

I was afraid I was using a lot of PRUs, but after seeing some of these, DAMN! Do you guy s just use a microphone and talk to the computer? Do you guys actually know how to code? Or do you just dictate instructions? or as they call it.. "Vibe code it"...

Haha, J/K

or am I?

[AlexPrincipal]

It’s over. How did it go for you guys?