GitHub Copilot Pricing Changes: What You Need to Know

My Copilot subscription just got nine times more expensive for the same work. Starting June 1st, the GitHub Copilot pricing changes wipe out the flat 300-request allowance I was paying $10/month for and replace it with a prepaid token meter that charges you the same per-token rate as buying Claude directly from Anthropic. The annual plan is gone too. If you were on it, like I was, it's not coming back.

I expected this kind of squeeze eventually. I did not expect it this fast.

What the new plan actually does

Until now, the deal was simple. Ten bucks a month bought you 300 premium requests. A request was a request. Claude Opus counted as a few of them, cheaper models counted as fractions, but the unit was a call, not a token count. You could send a 40k-token prompt and it cost the same one request as a one-line question. That made budgeting predictable. It also made Copilot a genuinely good deal if you used the heavy models a lot.

That model is being scrapped. Here is what replaces it:

• The Pro tier stays at $10/month and the Pro+ tier stays at $39/month, but your allowance is now a credit balance, not a request count.
• Credits get burned by tokens, priced to mirror the underlying provider (so Haiku output is around $5 per million tokens, same as Anthropic's API).
• The previously free models inside the IDE are no longer free against your plan.
• Code review, which runs on GitHub Actions under the hood, now bills you for those Action minutes too.
• The annual plan is being discontinued. Monthly only.
• Opus access is locked to the $39 tier.
• In-editor completions and edit suggestions remain unmetered. That's the one thing they're not charging extra for.

The only piece of the old offer that survives intact is code completion access in the editor. Everything else moved to a meter.

The model multiplier increase nobody is talking about

Here is the part that made me close the email and walk away from my desk for a minute.

GitHub kept the "premium request" concept around as a credits abstraction, but they reweighted what each model costs. The old multipliers were gentle. Claude Sonnet was 1x. Opus was 3x. Reasonable, given that Opus is genuinely more expensive to run.

The new multipliers are not gentle.

• Claude Sonnet: 1 → 9
• Claude Opus (older): 3 → 15
• Claude Opus (newer revision): 3 → 27
• Various mid-tier models: 1 → 3 or 1 → 6

That last one is a 9x jump on the same model. Not a price correction. A repricing.

Do the math on the $10 plan. If your 300 monthly requests now cost 27 credits each on Opus, you get roughly 11 Opus calls a month. On Sonnet at 9 credits, you get about 33. That is one Sonnet call per day. For anyone shipping real code, that runs out by lunch on Monday.

And you can't reach Opus from the $10 tier at all anymore. You have to be on Pro+ at $39.

GitHub Copilot pricing changes in plain numbers

Here is what the bill actually looks like once you spend past the included credits. Pricing per million tokens, matching the upstream provider:

• Haiku: ~$1 input, ~$5 output
• Sonnet: in line with Anthropic's published Claude API pricing
• Opus: same again, no discount

That is the part that broke it for me. There is no resale discount. You're prepaying for credits at the API's list price. If you blow through the prepay, you pay overage at that same rate. The middleman tax used to buy you predictability. Now you're paying the middleman tax and the API rate.

The old subscription gave you something the API never could: a hard ceiling. Hit your 300 requests, you stop. Pay $10, sleep at night. That ceiling is gone. The flat fee is gone with it. What's left is a prepaid wallet against metered usage, which is just an API account with worse UX.

Why this kills the value prop

The thing Copilot was selling, beyond the editor integration, was insulation from token-based billing. You did not have to think about prompt size. You did not have to compress context. You did not have to choose between Opus and Sonnet based on cost. You just used it.

A usage-based pricing model removes that insulation. Now every chat message has a price tag. Every long file you paste in costs more than a short one. Every reasoning-heavy model burns your balance faster. The mental overhead of "should I send this prompt" is back, and that overhead is exactly what people were paying $10 to avoid.

How much does GitHub Copilot cost per month now, really?

The sticker is still $10 or $39. The real cost is whatever your token usage adds up to, because almost nobody on a serious codebase will stay inside the included credits.

I ran a rough estimate against my own usage over the last 60 days. I average something like 8 to 12 chat interactions a day on a working day, mostly Sonnet, occasionally Opus for the harder refactors. Under the old plan that fit comfortably in 300 requests/month with room to spare. Under the new credit math, my same usage would burn through the $10 allowance in roughly the first week and the $39 allowance somewhere in the third week. After that I'm paying API rates on top.

So the honest answer to "github copilot premium requests pricing" is that there are no premium requests anymore. There are only tokens, priced at API parity, dressed up in a subscription wrapper.

Is GitHub Copilot worth it anymore?

For the in-editor completions alone? Maybe. They're still good, they're still unmetered, and the VS Code integration is the smoothest in the category. If you only use Copilot for ghost-text completions and never touch the chat panel, the $10 still buys you something the raw Anthropic or OpenAI APIs don't sell directly.

For everything else, the calculation has flipped. GitHub Copilot vs paying for Claude directly used to be an easy win for Copilot because of the request-based ceiling. Now it's a wash on price and a loss on flexibility. Claude's own subscription tiers (Pro at $20, Max at higher tiers) still offer something closer to a flat-fee experience with generous usage caps. ChatGPT Plus from OpenAI works similarly. Both give you a predictable monthly bill in a way Copilot's new structure does not.

If I were starting fresh today, I would probably:

• Keep Copilot Pro at $10 only for the editor completions
• Use a Claude Pro or Max subscription directly for chat, planning, and the heavy reasoning work
• Skip Copilot Pro+ entirely until the multipliers come down

That last point matters. Pro+ at $39 used to be a reasonable price for heavy users. With Opus now consuming 27 credits per call, $39 buys you somewhere around 15 to 20 Opus interactions before overage. That is not a developer tool, that is a tasting menu.

What this signals about the broader market

The interesting question is not whether Copilot specifically is still a good deal. It's whether any flat-fee AI subscription survives the next 12 months.

The economics here are brutal. Every provider, OpenAI included, is losing money on heavy users of unlimited plans. The original Copilot deal worked because most subscribers used a fraction of their 300 requests. The handful of power users who maxed out were subsidised by everyone else. As more users move from "occasional" to "all day every day," the subsidy math breaks.

So we get this: premium request removal, annual plan discontinuation, model multipliers that make heavy use uneconomic at the subscription tier, and a quiet shift to API pricing under a subscription label. None of it is technically a price hike. The sticker stayed the same. The thing behind the sticker got reshaped until paying for it makes the same financial sense as paying the upstream API directly.

I think Anthropic and OpenAI hold out longer on flat-fee plans because they own the model and have margin to play with. Resellers like GitHub do not have that margin, which is why Copilot moved first. But I would not assume Claude Max or ChatGPT Pro stay in their current form forever either. The direction is clear.

What I'm actually doing about it

Cancelling my Copilot Pro subscription before June 1st and downgrading my expectations. I'll keep an eye on whether the editor-only experience is worth $10 on its own. I'll move chat work to Claude directly because at least there I know what I'm paying for and the prepaid credit value matches the consumption rate.

If you've been on the annual plan, check your renewal date. The plan is going away regardless, but you want to know what you're being migrated to before the email shows up.

And if Copilot reverses any of this in the next 90 days because of the backlash, I'll happily come back. I just don't expect they will. Once a subscription business proves it can charge per-token, it doesn't usually go back to flat-fee. That door tends to close in one direction.