OpenTahr

AI economics

GitHub Copilot Token Billing and What It Does to Your Bill

Since June 1, every Copilot conversation is billed in tokens at the API rate of the model you pick. Developers' cost projections have jumped by an order of magnitude and the official discussion is now locked.

A hand dropping a coin into an old chrome parking meter on an office desk
The all-you-can-eat plan gives way to the meter. Every token Copilot uses is now counted. Illustration by OpenTahr.

01Five key points

  • Since June 1, 2026, every GitHub Copilot plan has moved to AI Credits. Usage is counted in tokens, at each model's published API rate.
  • Subscription prices stay the same ($10 to $39 per month), but included usage is capped at the plan's price. Beyond that, you pay for every token.
  • Code completions stay unlimited. Chat, agents and code review use up credits.
  • Developers are posting projections that go from $29 to $750 per month. Another goes from $50 to $3,000. The official discussion shows 958 downvotes against 24 upvotes and has been locked.
  • Promotional credits soften the transition until August 2026, with $30 per month free on Business and $70 on Enterprise.

0202What changed on June 1

Announced on April 27, the switch took effect on June 1, 2026. GitHub replaced its "premium requests" with GitHub AI Credits. One credit is worth $0.01. Each call to chat, an agent or code review uses credits based on the tokens exchanged (input, output and cache), at the published API rate of the model you choose. For example, Claude Sonnet 4.6 is billed at $3 per million input tokens and $15 per million output tokens, exactly as on Anthropic's API.

Plan prices don't change. What you get for them does. Each subscription now includes an amount of credits equal to its price, then switches to pay-as-you-go billing once that amount runs out.

Copilot plans after June 1, 2026, according to GitHub's official announcement.
PlanMonthly priceIncluded creditsFree until August
Copilot Pro$10$10none
Copilot Pro+$39$39none
Copilot Business$19 / user$19+$30 / month
Copilot Enterprise$39 / user$39+$70 / month

Two things stay off the meter. They are code completions and Next Edit suggestions, which every plan includes with no limit. Automated code review, on the other hand, now uses both credits and GitHub Actions minutes. Unused credits don't roll over from one month to the next. GitHub also confirmed that annual plans are being phased out and will move to the free tier when they expire.

Update of October 16, 2026. GitHub's official plans page now adds a flexible allotment on top of the base credits. Copilot Pro at $10 a month comes with $15 of AI Credits in total and Pro+ at $39 with $70, while a new Max plan at $100 a month includes $200. One credit is still worth $0.01 and code completions stay unlimited on every paid plan.

0303Bills that jump by an order of magnitude

Developers reacted right away. On the official discussion GitHub opened, the announcement post shows 958 downvotes against 24 upvotes and more than 530 comments. The thread ended up locked. TechCrunch built its headline on "What a joke", a phrase borrowed from one developer. Several users say GitHub changed the rules mid-game.

The figures going around show the scale. One developer who paid about $29 a month calculates that their current usage would cost nearly $750. Another posts screenshots going from about $50 to some $3,000. On the GitHub discussion, one user reports burning through 54% of a monthly quota in a single request on a small project. Microsoft had not responded to press inquiries when these accounts were published.

Be careful with these amounts. They are usage projections posted by the people concerned, not average bills. Developers who only use code completions will see almost no difference on their bill.

Ink drawing of a calm developer at a desk under a huge wave of coins
Agentic workflows burn through tokens by the million, so the wave is about to hit heavy users. Illustration by OpenTahr.

The people most exposed are the ones who rely on agents. When Copilot reads a codebase, plans changes, edits files, runs tests and starts over, each loop exchanges tens of thousands of tokens. That is exactly the use GitHub has been promoting in recent months. Freelance developers and small teams, with no budget cushion, take the hardest hit.

0404The end of a silent subsidy

For about three years, GitHub absorbed the gap between a flat fee and rising inference costs. That model held up as long as Copilot mostly served cheap code completions. The rise of agentic workflows, which multiply round trips to the models, made it unsustainable. The specialist press describes this as the end of years of silent subsidy.

Annual plan holders are spared AI Credits until their plans expire. Even so, their premium request "multipliers" jumped on June 1, from 1x to 9x for Claude Sonnet 4.6 and from 3x to 27x for Claude Opus 4.6. The message is the same everywhere. The price of AI is catching up with its real cost, meaning the API rates the providers charge.

One developer argued that whatever the marketing wrapping, a multiplier that goes from 1 to 9 is still a substantial price increase.

As quoted by Developpez.com, April 2026.

0505Keep control of your bill

If you stay on Copilot, four habits limit the damage. The first two are about spending limits and model choice. The last two are about watching your usage and making the most of what stays free.

  • Set caps. Organizations can set budgets per user, per group or per cost center. Do it before the promotional credits end in August.
  • Pick the model for each task. An agent on a big project justifies a top-tier model. A one-off question works fine with an entry-level model that costs several times less per token.
  • Watch your usage. The billing dashboard breaks down usage by model. The developer who burned half a quota in one request shows what a poorly scoped agent can cost.
  • Make the most of what stays unlimited. Completions and Next Edit cover a large share of daily work without touching your credits.

The specialist press also reports that the context window in Copilot is capped at 160,000 tokens, even for models that accept more when called directly. For very large contexts, the direct API keeps a technical edge on top of the price advantage. That gap counts most when an agent has to load a large codebase in one go.

0606What this says about the market

Copilot is not an isolated case. It signals that the all-you-can-eat AI subscription does not survive the arrival of agents. When the middleman bills tokens at the provider's API rate, its value moves. It no longer comes from subsidizing compute but from the tooling around it.

OpenTahr has been built on this conviction from day one. You might as well plug in your own API key and pay the provider the real price, with no layer in between. The alternatives that angry developers mention point the same way, from Anthropic's or OpenAI's direct API to aggregators like OpenRouter and local open source models. You can read how BYOK keys work on the home page. The advice we gave about Claude Fable 5 holds more than ever, so set a spending cap with your provider before you launch long agentic tasks.

Ink drawing of a hand holding a key before a row of open doors
When the middleman charges API prices, a direct key becomes the shortest path again. Illustration by OpenTahr.

For web professionals, the lesson goes beyond Copilot. Before you build a workflow on an AI subscription, think about what happens the day the vendor aligns its prices with its costs. A tool that runs on your own key, at your provider's rate, can't pull that trick on you. The bill comes from the provider itself, with a cap you set.

0707Frequently asked questions

Is the GitHub Copilot subscription price going up?

No. Plans keep their prices, from $10 (Pro) to $39 (Pro+ and Enterprise). What changes is what they include, because each plan now comes with credits equal to its price. Once those credits run out, extra usage is billed on top.

What stays unlimited?

Code completions and Next Edit suggestions stay unlimited on every plan. Chat, agents and code review work differently. They use up AI Credits calculated in tokens, at the API rate of the model you pick.

Who is hit hardest?

The people hit hardest are users of agentic workflows, where Copilot works on its own across the codebase. These sessions exchange huge numbers of tokens. Freelancers and small teams are the most exposed, since they have no budget cushion.

How do you avoid a surprise bill?

Set budget caps in your organization settings and track usage by model. Keep expensive models for the tasks that justify them. Also use the free credits offered until August 2026, which come to $30 per month on Business and $70 on Enterprise.

0808References

  1. GitHub, "GitHub Copilot is moving to usage-based billing", official announcement, April 27, 2026.
  2. GitHub Community, official discussion on usage-based billing, accessed June 10, 2026.
  3. TechCrunch, "'What a joke': GitHub Copilot's new token-based billing spurs consternation among devs", May 30, 2026.
  4. Blog du Moderateur, "GitHub Copilot switches to usage-based billing" (in French), June 2, 2026.
  5. Developpez.com, "GitHub Copilot drops unlimited use and moves to usage-based billing on June 1" (in French), April 28, 2026.
  6. Geeek.org, "GitHub Copilot token billing from June 2026" (in French), May 12, 2026.
  7. GitHub, official Copilot plans page, accessed June 10 and October 16, 2026.

Want to try these AI models on your own questions, with your own key?

Explore the OpenTahr tools
Illustrated portrait of Anthony

Anthony

Founder of OpenTahr

A French entrepreneur who builds online products around the web, SEO and AI. The methods in OpenTahr are the ones he uses on his own sites.