GitHub Copilot Kills Unlimited Plans for AI Credits

The flat-rate era of AI coding assistants ends on June 1. GitHub is replacing Copilot's premium request allowance with a metered currency that charges you for every input, output, and cached token your model touches. Developers ran the math and reached an uncomfortable conclusion: same monthly price, far fewer requests.

What Actually Happened

GitHub confirmed that on June 1, 2026, all Copilot plans move from premium request units to usage-based billing denominated in a new virtual currency called GitHub AI Credits, priced at $0.01 each. Credits are consumed based on token usage, including input, output, and cached tokens, billed at each model's published API rate. A query to an expensive frontier model now costs many multiples of a query to a cheap one, and the meter runs on every token rather than every request.

Each plan ships with a monthly credit allotment that mirrors its price. Copilot Pro+ at $39 per month includes $39 in credits, Copilot Business at $19 per user includes $19 in credits, and Copilot Enterprise at $39 per user includes $39 in credits, with paid plans able to buy more. Monthly subscribers convert automatically on June 1, while customers on annual Pro or Pro+ plans keep request-based pricing until their term expires. GitHub framed the shift as a response to escalating inference costs, where complex agentic coding sessions made the old unlimited model financially unsustainable.

Why This Matters More Than People Think

This is the moment AI coding stopped being a subscription and became a utility bill. Under request-based pricing, a developer could fire off a thousand prompts a month for a fixed fee, and the heavy users were subsidized by the light ones. Token metering ends that cross-subsidy. The engineer who runs long agentic sessions with large context windows now pays for the compute they actually consume, and early reactions suggest that for power users the same $39 buys meaningfully less work than it did under the old plan.

The deeper shift is psychological. A flat fee invites experimentation: you try things because the marginal cost is zero. A meter does the opposite. Every developer who has watched a cloud bill spike knows the instinct to ration, and that instinct is now wired into the daily act of writing code. GitHub is betting that the convenience of Copilot outweighs the friction of the meter, but it is introducing cost-anxiety into a workflow that previously felt free at the point of use, and that changes how people use the tool.

The Competitive Landscape

GitHub Copilot does not operate in a vacuum. Cursor, backed by a $2 billion raise at a reported $50 billion valuation, has built its brand on aggressive AI coding features and its own pricing experiments. Anthropic's Claude Code and OpenAI's Codex compete directly for the same agentic-coding budget, and both already meter by token at the API layer. By moving Copilot onto a token meter, Microsoft is aligning Copilot's economics with the rest of the market, but it is also surrendering the one structural advantage it had: predictable, all-you-can-eat pricing that enterprises could budget against.

The competitive risk is defection at the margins. A developer who feels nickel-and-dimed by credit consumption has never had more alternatives, and switching costs for coding assistants are low because the editor integrations are increasingly standardized. Cursor and Claude Code will market themselves as the place where you can still think without watching a meter, even if their underlying economics are similar. Microsoft's counter is bundling: Copilot ships inside GitHub, VS Code, and the broader Microsoft 365 estate, and that distribution moat is harder to switch away from than a standalone tool. The question is whether bundling holds users who feel the pricing turned against them.

Hidden Insight: The Unit Economics Finally Showed Up

The non-obvious story is that GitHub did not change its pricing because it wanted to. It changed because the unit economics of agentic coding broke the subscription model. When Copilot was autocomplete, a request was cheap and predictable. Now that Copilot runs multi-step agents that read entire repositories, hold long context, and call models repeatedly to complete a task, a single session can burn through dollars of inference. Unlimited pricing only works when the marginal cost of a request is trivial. Agents made it anything but trivial, and the subscription math stopped closing.

This is the first visible crack in a pattern that will spread across the entire AI software industry. Every SaaS company that bolted an AI feature onto a flat monthly fee is sitting on the same time bomb: the more valuable the AI feature, the more compute it consumes, and the more it erodes the margin of a fixed-price plan. The vendors that win the next two years will be the ones who figured out metering before their best customers became their least profitable ones. GitHub is simply early to admit it out loud.

The uncomfortable truth this challenges is the assumption that AI software follows the economics of traditional software. Classic SaaS has near-zero marginal cost, which is why flat subscriptions work and gross margins hit 80%. AI features have real, variable marginal cost that scales with usage, which makes them behave like infrastructure, not software. The companies still pricing AI like SaaS are, in effect, running a business where their most engaged users are their biggest losses. Token metering is not GitHub being greedy. It is GitHub admitting that the old model was structurally underwater, and that the rest of the industry will have to follow whether customers like it or not.

What to Watch Next

In the first 30 days after June 1, watch the GitHub community forums and churn signals: the volume and tone of complaints about credit burn rates will tell you how badly the change hit power users. Watch whether GitHub adjusts the included allotments, because a quiet bump in monthly credits would signal the initial caps were set too aggressively. In the next 90 days, track Cursor and Claude Code marketing for explicit anti-meter messaging, which would confirm that pricing friction is driving real switching behavior.

Over 180 days, the metric that matters is whether other AI-feature vendors follow GitHub onto token metering. If Notion, Salesforce, or Atlassian quietly introduce credit systems for their AI features, that confirms the subscription-to-utility shift is industry-wide rather than GitHub-specific. The risk, however, is real and worth tracking: if developers respond by rationing their AI usage, the productivity gains that justified the tools in the first place could shrink, and a meter that makes engineers hesitate before asking for help may quietly undercut the entire value proposition. Critics argue this is the worst possible outcome, where the pricing model trains users to use the product less.

AI coding just stopped being a subscription and became a utility bill, and your most engaged developers will feel it first.

---