Starting June 1, 2026, GitHub Copilot will switch to token-based billing, moving away from the current Premium Requests system, where each query counts as a request regardless of its complexity. Under the new model, charges will be based on the number of tokens processed, and fees will depend on the model chosen, input and output volume, cache size, and features used.
Subscription prices will remain unchanged. Instead of a fixed number of queries each month, users will receive AI credits at the same price. For example, a paying Copilot Pro subscriber $10 one month you will get 1000 AI creditsand GitHub values each credit at a cent.
How GitHub Copilot token-based billing will work
One token is approximately three-quarters of a word. A codebase containing 10,000 expressions, declarations, and variable names typically equals 12,000 to 13,000 tokens. Both input and output cues from the model contribute to token usage. The cost per credit varies depending on the model.
Using more advanced models consumes credits more quickly than less capable ones. Tasks involving large codebases with multiple agents will use credits at a faster rate than simple single-shift queries. Code completion and next edit suggestions are not affected by the new billing model and will remain free.
What the new GitHub Copilot billing means for users
Under the previous model, users could typically use three to eight times the token value covered by the cost of their subscription, with GitHub covering the difference. This subsidy will end on June 1. Users who primarily run simple queries are unlikely to go over their monthly credit limit, but those who run more complex workflows on large codebases may need to purchase additional credits.
GitHub has not published a breakdown of fees per token for each supported model level.
Why GitHub is moving Copilot to token-based pricing
The change aligns Copilot with pricing adjustments already implemented by Anthropic and OpenAI for your enterprise customers. Token-based billing gives AI providers a more direct view of actual computing costs, but also makes billing less predictable for users who previously paid a flat monthly fee regardless of usage.
