GitHub will start charging Copilot users based on their actual AI usage

GitHub says it can no longer absorb "escalating inference cost" from it heaviest AI users.

See full article...

 

[Aurich]

I changed my Creative Cloud plan with Adobe to a lower tier to avoid their AI bullshit, so I forget exactly what the new price I'm paying it now. I think they forced me into a monthly instead of annual.

But it's $600ish a year I think. I feel like the AI level I dumped was $800ish.

Similar tools for professionals are on that level. You can pay about that for Fusion 360's full CAD package for instance.

It's a normal and basic cost for doing business. I don't love subscriptions, but I'll pay them to do my work the way I want/need to do it.

I don't use LLMs, but looking at Claude their best plan is $100 a month, so double my Creative Cloud costs. And okay, again if you're a professional there are costs for getting your work done you just pay.

I guess the question is simple: is $100 enough? Is there a future where pros just pay that kind of cost and use the tools and it's just part of doing business?

Or is it not sustainable still at that cost? Or are there not enough people to pay it?

At this point it feels like almost a best case scenario that people who have genuine use cases for these tools just pay to use them, and the slop dries up because the free spigots get turned off and the people who can't justify it move on.

 

[Xyler]

Just the basic idea that all code in the world would be generated by 2 or 3 companies' models which require massive data centers to run is so objectively ludicrous. Did none of these people think through the logical conclusion ?

No, these are the same people who think having a Cloud PC Rental Service is gonna work super well. Jeff Bezos mentioned this idea around, and I cannot even begin to laugh at how terrible such a project would go.

 

[Fred Duck]

It's the same concept that all subscription services use. They bet on a lot of users who don't use it at all, and many who don't use it much to pay for the people who use it constantly.

CtCoHUwMoLU it is!

I can't wait to see people pop into articles, simply post that then dash off, and garner hundreds of upvotes! It will add much value to conversations!

 

[SteveBigandTall]

YES!
Folks having to pay for usage means one thing.
The bubble is gonna burst by August 01, October 01 at the latest.

 

[caramelpolice]

This is insane cope: these price increases are possible because corporations have ventured beyond chatbots and are now pursuing agentic flows at scale. Companies are finding immense value in this and are willing to pay the steep premium for frontier models.

If enough companies were actually willing to pay a 'steep premium' then there would be no reason for Microsoft (or anyone else) to be losing a single penny on these services.

 

[LDA 6502]

Meanwhile, my workplace just sent out a warning about minimum LLM usage in order to retain a license. This is on top of a warning the other month about over-usage.

So, use it, but not too much. "No, not like that!"

 

[AdamWill]

Klarna rehired human customer service reps after firing them all for AI (but they are being offered gig worker jobs).

My team has reported 3 or 4 major GitHub outages in the past two weeks. We’d pay extra for GitHub classic with all the old fashioned uptime and none of the new AI “features.”

Edit for typo

Yesterday the freaking issues list wasn't listing all the issues. How do you break that? In production?!

 

[Tactical Finesse]

I changed my Creative Cloud plan with Adobe to a lower tier to avoid their AI bullshit, so I forget exactly what the new price I'm paying it now. I think they forced me into a monthly instead of annual.

But it's $600ish a year I think. I feel like the AI level I dumped was $800ish.

Similar tools for professionals are on that level. You can pay about that for Fusion 360's full CAD package for instance.

It's a normal and basic cost for doing business. I don't love subscriptions, but I'll pay them to do my work the way I want/need to do it.

I don't use LLMs, but looking at Claude their best plan is $100 a month, so double my Creative Cloud costs. And okay, again if you're a professional there are costs for getting your work done you just pay.

I guess the question is simple: is $100 enough? Is there a future where pros just pay that kind of cost and use the tools and it's just part of doing business?

Or is it not sustainable still at that cost? Or are there not enough people to pay it?

At this point it feels like almost a best case scenario that people who have genuine use cases for these tools just pay to use them, and the slop dries up because the free spigots get turned off and the people who can't justify it move on.

Guessing that would be a no:

https://meincmagazine.com/ai/2026/04/anthropic-tested-removing-claude-code-from-the-pro-plan/

That was the $20 plan...but odds are with increased demand for the $100 plan, it would lose viability too.

 

[Tactical Finesse]

NVIDIA isn’t saying that they spend more on compute than employees as a signal that costs are out of control. They are saying it in the sense of “the FANUC robots in our factory cost 10x what we pay a worker”

Knowledge workers are heavily leveraging AI and even a couple thousand dollars in compute cost per worker per month is nothing compared to the fully loaded cost of an incremental employee.

That depends on how large a blast-radius of lost productivity cleaning up after the sycophantic hallucinating LLM is. Law firms are finding out first hand right now.

 

[Aurich]

Guessing that would be a no:

https://meincmagazine.com/ai/2026/04/anthropic-tested-removing-claude-code-from-the-pro-plan/

That was the $20 plan...but odds are with increased demand for the $100 plan, it would lose viability too.

Pretty big difference between $20 and $100 so it's hard to know.

 

[msawzall]

Don't use AI and don't pay! Follow me for more money saving tips.

 

[Pluvia Arenae]

YES!
Folks having to pay for usage means one thing.
The bubble is gonna burst by August 01, October 01 at the latest.

Markets can stay irrational longer than you can maintain your excitement about a bubble bursting.

 

[msawzall]

Meanwhile, my workplace just sent out a warning about minimum LLM usage in order to retain a license. This is on top of a warning the other month about over-usage.

So, use it, but not too much. "No, not like that!"

Only use it for important stuff. Like ordering lunch for the c suite.

 

[uggrams]

Klarna rehired human customer service reps after firing them all for AI (but they are being offered gig worker jobs).

My team has reported 3 or 4 major GitHub outages in the past two weeks. We’d pay extra for GitHub classic with all the old fashioned uptime and none of the new AI “features.”

Edit for typo

The github status page is one of my most frequently visited pages in the last couple months. They have outages that interrupt my or a co-worker's workflow at least a couple times a week.

 

[stormcrash]

Don't use AI and don't pay! Follow me for more money saving tips.

Sadly my work basically requires that I do right now. It actually does make some things a lot less hassle, but I hate to think about how many trees I'm burning daily to "be productive" in the current bubblescape

 

[J.King]

Yesterday the freaking issues list wasn't listing all the issues. How do you break that? In production?!

Pull requests are still not listing quite properly, either. The Web Platform Test repository is missing half its PRs, still, for example.

 

[momoisdabest]

I changed my Creative Cloud plan with Adobe to a lower tier to avoid their AI bullshit, so I forget exactly what the new price I'm paying it now. I think they forced me into a monthly instead of annual.

But it's $600ish a year I think. I feel like the AI level I dumped was $800ish.

Similar tools for professionals are on that level. You can pay about that for Fusion 360's full CAD package for instance.

It's a normal and basic cost for doing business. I don't love subscriptions, but I'll pay them to do my work the way I want/need to do it.

I don't use LLMs, but looking at Claude their best plan is $100 a month, so double my Creative Cloud costs. And okay, again if you're a professional there are costs for getting your work done you just pay.

I guess the question is simple: is $100 enough? Is there a future where pros just pay that kind of cost and use the tools and it's just part of doing business?

Or is it not sustainable still at that cost? Or are there not enough people to pay it?

At this point it feels like almost a best case scenario that people who have genuine use cases for these tools just pay to use them, and the slop dries up because the free spigots get turned off and the people who can't justify it move on.

Presumably $100/month won't provide enough tokens to meaningfully generate much code, seeing as they were recently experimenting with throttling or whatever recently. They will all go to pay as you go, but to actually generate the amounts of code needed to "10x" entire programs' software output will be a prohibitive cost.

On a side note, just imagining the amount of energy that goes into this (resources and carbon footprint for chips, data centers, training, inference) and the output you get - code for ads or whatever crap features/apps these companies are developing just feels insane to me

 

[UhtredsonofUhtred]

This should be entertaining!

I'm guessing a lot of people have been using Copilot precisely because the cost has been subsidised. So those people will now jump ship to whatever LLM provider is still offering a subsidised service. And then that LLM provider will have to raise their prices...

And then the industry will go around this loop a few times, until there's no more subsidised LLM services left for people to take advantage of. At which point, we'll finally see how big the commercial customer base actually is for LLM models.

I suspect the results will be far too small to justify the current valuations of companies such as OpenAI!

They'll just start/increase the use of China's open-source models, which are "good enough" now and will likely reach parity soon.

OpenAI will be the pets.com of the AI bubble. With multiple orders of magnitude more tears.

 

[HeadPlug]

For the tool in question: you can select models from several providers without friction and GitHub is passing through API usage at retail cost for enterprise users. As APIs change price, copilot users should be able to quickly adapt to model pricing changes.

What's in it for GitHub? At some point, the middleman will start asking for rent/'commission'

 

[rojcowles]

I hear many anecdotes about individual employees burning $10k+/month of tokens for marginal gains in productivity. Look up "tokenmaxxing" if you want to melt your brain. GitHub probably made this change because they can't viably sustain a bunch of whales lobsters on a subscription model.

Right now tokenmaxxers are getting kudos, bonuses, and promotions for helping their employers' AI adoption metrics. I can't help but wonder what will happen to them once the model providers start needing to turn a profit.

Starting to wonder if one of the big draws of pushing AI everywhere for the senior execs in tech companies is that it gives them something trivially easy to measure, number of tokens burned by devs, that they can ASS-U-ME is in some way correlated to actual productivity and delivery of fixes and features to customers.

Whether that's true or not I can imagine the siren song of having the "One true number" to look at on a dashboard every hour or so without any need to understand what developers do, how customers use the product or how revenue is being impacted by the flood of AI driven code changes.

Meanwhile I'm off to get a job in construction where I hear Nail-maxxing is the hot new trend. BRB just need to find a random sheet of 4-by-8 I can fire a couple thousand nails into to meet my daily quota, top the leader-board and head home with a huge bonus.

 

[silverboy]

The funny thing to me, or the galling thing, is that they word their announcement as though people have somehow been abusing their LLM privileges. "Hey you bad kids, why didn't you keep it under control?"

This is what you wanted, jackasses, remember? So nauseating.

Beyond that, everyone else here has already said what there is to say about this bullshit.

 

[MilanKraft]

What's that I hear - is it the sound of the ice starting to crack under the feet of the managers who insisted that they could get rid of humans and just use LLMs instead?

This mess can't break up and sink careers fast enough.

Why stop at managers? Eagerly awaiting news accounts of at least a few CEOs / CIOs / CTOs getting shit-canned just in time for 4th of July. I've seen enough old-school fireworks in my day; I want to see corporate ones, preferably implosions.

 

[Dark Jaguar]

Copilot is for entertainment purposes only, according to Microsoft.

Apparently, that also means it's not for profiting off of either.

Also, Microsoft did the Microsoft thing when it came to their promise to scale back Copilot in every single program in Windows. They "removed" the name, only, from Notepad. It's still THERE, it's just now called "Writing tools", with the little fairy dust sparkle thing that I guess we were all supposed to naturally figure out is the symbol for "LLM chatbot"? I'm getting rather sick of them foisting these weird symbols on us, with NO name or explanation of their function, just expecting us to find all that out on our own because it's just "so obvious".

It's ALRIGHT to make symbols that look like real world objects! No, not the floppy disk one. That's the ONE symbol you SHOULD find something else for! The gear icon for settings? We still USE gears, it's fine! Arrows? Archery is still a popular sport, that too is fine.

 

[s73v3r]

This is insane cope: these price increases are possible because corporations have ventured beyond chatbots and are now pursuing agentic flows at scale. Companies are finding immense value in this and are willing to pay the steep premium for frontier models.

Absolutely fucking not. They are doing this because so far companies are not finding immense value in this. These things are super fucking subsidized, and no company is making any money with AI, Nvidia being the lone exception.

 

[s73v3r]

Cloud providers are enjoying continued strong margins for compute, they are not losing money at all.

Prove it. Microsoft is a public company. Find the last quarter report where they showed that AI compute was making money.

Model providers are in a deep cash burn to acquire users while simultaneously racing towards training the next model, and the API costs seem to be breakeven at least on gross margin, even if the operational cash burn behind it is unsustainable.

No, they are nowhere close to breakeven

 

[hanharal]

Free is always going to be popular but "Do I want to spend $1 to have this thing re-write an email for me?" is a whole 'nother thing.

Based on the new pricing and using GPT 5.4 mini, an email would cost something like 2-6 cents depending on length.

 

[42Kodiak42]

I got a feeling this is going to put a damper on AI usage way more than just the economics of cost and utility work out, simply because users are still charged for when the AI gets it wrong.

All those hallucinations and AI-generated errors are going to be incredibly frustrating when you pay for wrong answers and have to pay extra for the AI to try to correct them. Humans (when they're not deceiving you) are at least making progress towards working code when they get things wrong, and usually have a good upper bound of how much more time and money it'll cost to see it through. You won't get that with AI.

 

[gosand]

So how long before the article about someone who left their agent running for several days looking for the seahorse emoji and now they have a $10k bill they can't pay?

Yep, this smells a lot like the days when everyone had to get to the cloud at all costs... then they saw that first bill. I still remember the conversations
"I thought we could spin down and spin up VMs as needed based on workload?"
"Yes, that's possible, but we had 3 months to move this application to the cloud, so we did a lift and shift"
"What would it take to do it the right way?"
"We would have to rearchitect it, after we learn how to rearchitect it"

 

[42Kodiak42]

I changed my Creative Cloud plan with Adobe to a lower tier to avoid their AI bullshit, so I forget exactly what the new price I'm paying it now. I think they forced me into a monthly instead of annual.

But it's $600ish a year I think. I feel like the AI level I dumped was $800ish.

Similar tools for professionals are on that level. You can pay about that for Fusion 360's full CAD package for instance.

It's a normal and basic cost for doing business. I don't love subscriptions, but I'll pay them to do my work the way I want/need to do it.

I don't use LLMs, but looking at Claude their best plan is $100 a month, so double my Creative Cloud costs. And okay, again if you're a professional there are costs for getting your work done you just pay.

I guess the question is simple: is $100 enough? Is there a future where pros just pay that kind of cost and use the tools and it's just part of doing business?

Or is it not sustainable still at that cost? Or are there not enough people to pay it?

At this point it feels like almost a best case scenario that people who have genuine use cases for these tools just pay to use them, and the slop dries up because the free spigots get turned off and the people who can't justify it move on.

I don't think that the economics of AI usage are going to work out very well. In my previous comment, I mentioned that paying for tokens seems like it would turn into a frustrating, grating experience as you pay varying amounts for wrong answers without knowing how much it's actually going to cost you.

I don't think the "all you can eat buffet" pricing will work out because AI users have insanely different use-case profiles, even as you get to "vibe-coders" (I.e., someone who vibe-codes a module once a week vs. someone who's constantly prompting an coding AI for 8 hours a workday).

And I think a lot of the major sales power that drives Adobe and Fusion subscriptions starts to dry up when you get to the interchangeable arms race of AI models. Adobe software represents an entire workflow that you would need to replace, which becomes very difficult to do mid-project and gets in the way of collaboration with your coworkers if you're not unified in this change. The price being charged for AI models is just for one component that's, at most, a plugin away from being replaced, or just dropped for months on end as the user concludes "I don't really need AI tools for anything other than steps 21 and 43 out of 500"

 

[murty]

I work for a school, and we’ve been seeing these kinds of changes coming down the pipe for the last month or two on various “AI” products that are either part of stuff we already use (Office 365 and Google Suite), as well as some 3rd party tools used by some employees for various reasons.

We (IT) have been doing our best to gatekeep all this crap as best as we could, but there are political (not technical) reasons why we’ve been shoved more into this stuff than we would prefer to have on our network.

We’ve been warning that the other shoe would drop soon, regarding pricing, but no one listened. And now we’re starting to see their pricing models shift to funny money tokens, which have the added “benefit” or obscuring costs and making them incredibly hard to futurecast budgets for. I’m sure that part of this tokenization plan is to do exactly that, abstract the costs in a way that makes it difficult to mentally understand usager rates vs token costs.

Anecdotally, we’ve been using Asana for various things at our school. The other day, someone turned on some kind of LLM powered tool for their business office workflows. They barely set things up and maybe ran it a few dozen times before our pool of tokens was rapidly drained. Our entire initial allotment gone in > 30 minutes. Asana wanted to bump us to a higher license tier to increase our token pool. This is going to get really “fun” soon.

Probably going to have to start using similar tools to AWS billing estimator calculators in order to properly budget this dumb garbage.

This is about to completely wreck everyone’s budgets (and not just schools). Good luck to the idiots that shifted substantial portions their business model to “AI”, I suspect you’ll wish you had retained more employees/institutional knowledge because you trimmed your headcount.

The free lunch was always going to end eventually, those of us who weren’t drinking the Kool-aid have been warning about this kind of thing for a while, but I suspect a lot of less aware folks are in for a nasty surprise soon.

More and more these days, I am seriously considering a career change because of where this all seems to be going, and I want no part of it. But, I have no what I could possibly shift to in my mid 40s that would keep a similar rate of pay.

Edit: also, can’t help but think of this Simpsons bit:
View: https://youtu.be/K-6IzkDyl9Q

 

[hanharal]

For us not using these tools, what is a million tokens enough for? I understand it being around a million words, which is quite a lot. If you write 4 a4 pages per day that is around 2 years of output. Now, I understand you cannot just say those are equal, just single shot slop is not the same as targeted human output

It's very hard to estimate.

Casual User (Personal Use): 5,000 – 15,000 tokens per day
Asking 3–5 brief questions, requesting a summary of a short text, or drafting an email.

Moderate User (Office / Administration): 20,000 – 100,000 tokens per day
Equivalent to: Reviewing and summarising reports, engaging in multiple extended conversations where context (history) accumulates and generating drafts for articles or documents.

Power User (without coding): 100,000 – 500,000+ tokens per day
Equivalent to: Uploading large PDF files (100+ pages) into models and querying them. This consumes large amounts of tokens because the entire document must be re-read for every follow-up question.


You have this snowball effect where the system has to read the entire conversation for every new question or request. If you have a chat with 20 exchanges back and fort, you can easily consume 10,000 tokens.

 

[murty]

For us not using these tools, what is a million tokens enough for? I understand it being around a million words, which is quite a lot. If you write 4 a4 pages per day that is around 2 years of output. Now, I understand you cannot just say those are equal, just single shot slop is not the same as targeted human output

Now, these more useful tools burns a lot more tokens, but how many? Say you needed to write C code to handle the SENT protocol ( wikipedia link ) on a MCU. How many tokens would you burn to write and test that?

I think their hope is that most people don’t understand it, so they can rip them off more.

 

[DarthSlack]

The stock market thinks Microsoft is in trouble.

The stock market also thinks that Tesla is the bestest company evar in the history of companies.

The stock market is not bright.

 

[tim305]

Ah yes, if there is one thing I expect from my employees it's the financial training and competence to maximize their marginal utility. That's why every one has signing authority.

https://en.wikipedia.org/wiki/Homo_economicus

Yeah, I guess it's pointless to put meters on electricity and gas pumps.

 

[stormcrash]

We’ve been warning that the other shoe would drop soon, regarding pricing, but no one listened. And now we’re starting to see their pricing models shift to funny money tokens, which have the added “benefit” or obscuring costs and making them incredibly hard to futurecast budgets for. I’m sure that part of this tokenization plan is to do exactly that, abstract the costs in a way that makes it difficult to mentally understand usager rates vs token costs.

This is especially true since it's not 1 prompt = 1 token, you have no way to know beforehand or control how many "tokens" are going to be consumed to process your prompt/request.

It's a complete black box, and if the token becomes the billable unit then the companies begin to lose the incentive to reduce token consumption and make the models more efficient because that would mean users need fewer tokens and then will spend less,

 

[eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9]

paying for tokens seems like it would turn into a frustrating, grating experience as you pay varying amounts for wrong answers without knowing how much it's actually going to cost you

I agree, but this is also the experience with humans too, yet with a slower iteration loop.

AI software engineering tools don’t have to be perfect to be valuable. Even if the cost rises considerably.

I don’t know a single dev who doesn’t think the game has changed, permanently. My team has been noticeably more impactful since ~Nov ‘25 (Opus release).

GH Copilot’s strategy? I’m more iffy on that. This 100% feels like an enterprise squeeze.

 

[J.King]

This is especially true since it's not 1 prompt = 1 token, you have no way to know beforehand or control how many "tokens" are going to be consumed to process your prompt/request.

It's a complete black box, and if the token becomes the billable unit then the companies begin to lose the incentive to reduce token consumption and make the models more efficient because that would mean users need fewer tokens and then will spend less,

I'm sure they'll "accidentally" play some token accounting tricks before too long. Telecomm companies would kill to have a metric like that which the customer can't measure themselves.

 

[stormcrash]

I'm sure they'll "accidentally" play some token accounting tricks before too long. Telecomm companies would kill to have a metric like that which the customer can't measure themselves.

It was called "message units" back in the day. In large cities local calls were billed message units based on distance and time, and unlike long distance those local calls were not itemized. Not even the phone phreaks could consistently track their message unit usage reliably.