GitHub Copilot's Flat-Rate Era Ends as Token Billing Shocks Developers

GitHub just ended one of the best deals in developer tooling. As of June 1, 2026, GitHub Copilot moved from flat-rate Premium Request Units (PRUs) to token-based billing called GitHub AI Credits. The subscription prices haven't changed, but what you get for that price has. And if you run agentic coding sessions, the difference is not subtle.

Zed was among the first editors to flag the impact directly to its users. Copilot Chat is now metered with GitHub AI Credits, while Copilot edit predictions (code completions and Next Edit suggestions) are not. This applies anywhere you choose a Copilot Chat model in Zed: agent turns, inline assists, commit-message generation, tool calls, and subagent work all count. If you use Copilot Chat heavily, especially with agents, you should expect this to cost more.

From gym membership to pay-per-rep

The old model was simple: pay a flat monthly fee, get a bucket of premium request units, and each model interaction costs one unit with a multiplier based on model power. Before June 1, billing was premium request-based: each model interaction cost one premium request unit, and a multiplier was applied based on which model you used. The new model works completely differently.

Usage is now calculated based on token consumption, including input, output, and cached tokens, using the listed API rates for each model. The conversion is fixed: 1 AI Credit equals $0.01 USD, so a model that costs $30 per million output tokens burns 3,000 credits for a million tokens of output. The key insight is that a quick chat question and a long autonomous agent session no longer cost the same. Usage now scales with the actual compute a request triggers, which makes heavy agentic workflows far more expensive than they were under the flat PRU model.