GitHub Copilot Now Supports 1 Million Token Context and Configurable Reasoning

GitHub added two new capabilities to Copilot on June 4: a one-million-token context window and configurable reasoning levels. Both are available now in VS Code, the Copilot CLI, and the GitHub Copilot app, with more surfaces coming.

The context window change

Working across large codebases has been a persistent friction point with AI coding tools. Context runs out before a refactor is done, or you end up manually chunking work to fit within limits. The expanded window addresses that directly.

With models like Claude Sonnet 4.6 and Claude Opus 4.8, the 1 million token window lets Copilot hold a much larger slice of your codebase at once. That’s enough to include many related files, their tests, and relevant configuration without dropping context mid-task.

The tradeoff is credits. Using the extended context window consumes more AI credits per interaction. GitHub recommends reserving the larger context for complex, multi-file work rather than routine questions. For quick lookups or single-file edits, the standard window is more economical.

Configurable reasoning levels

Reasoning level controls let you choose how much processing depth the model applies before responding. Higher levels are meant for hard problems: architectural decisions, complicated debugging, and tasks where accuracy matters more than speed. Lower settings give faster responses for simpler requests.

Same tradeoff applies: deeper reasoning costs more credits.

Context in the June 1 billing shift

Both features make more sense in light of the usage-based billing that went live on June 1. With per-credit pricing, GitHub can offer premium capabilities without raising base plan prices. The cost scales with how much you use.

That’s useful for developers who need extended context occasionally but not constantly. You pay for the extra capacity when you actually need it, and you don’t pay for it on the queries where you don’t.

Whether the credit math works out favorably depends on how you use Copilot. For heavy agentic work in large repos, 1M context could meaningfully reduce the number of failed or truncated agent sessions.

Source: GitHub Changelog, June 4 2026