Skip to content
News Product launch

GitHub Copilot adds 1-million-token context windows and configurable reasoning levels

· by Pondero Newsdesk

The short version

GitHub Copilot now supports a 1-million-token context window and user-selectable reasoning levels in VS Code, Copilot CLI, and the Copilot app. Both features draw from AI Credits under the usage-based billing model that went live June 1.

GitHub Copilot adds 1-million-token context windows and configurable reasoning levels

GitHub published a changelog entry on June 4, 2026 adding two capabilities to Copilot: a one-million-token context window and configurable reasoning levels. Both are available now in VS Code, Copilot CLI, and the Copilot app, with a rollout to additional surfaces described as coming soon.

What

The one-million-token context window lets developers load large codebases, lengthy design documents, and complex multi-file projects into a single Copilot chat session without hitting a context ceiling mid-session. Prior to this update, Copilot sessions worked within much smaller context limits tied to the underlying model's default window, which required developers to break large tasks into smaller chunks or manually curate what files the model could see.

Per the June 4 GitHub changelog entry, the feature is available today in VS Code, Copilot CLI, and the GitHub Copilot app. GitHub described expansion to more surfaces as coming but did not specify a timeline.

The configurable reasoning levels feature lets developers choose how much thinking effort Copilot applies before producing a response. A lower reasoning level returns answers faster and uses fewer AI Credits per interaction. A higher level triggers extended thinking suited to architectural decisions, complex bug diagnosis, and multi-repository refactors. GitHub described the dial as a speed-versus-depth tradeoff, per the same changelog entry.

Both features increase AI Credit consumption relative to standard Copilot requests. Per the June 4 changelog, GitHub recommends sticking with the default context window and default reasoning level for everyday tasks and reserving the extended options for complex, multi-file problems. No specific credit rate for either feature was published in the changelog. The model documentation linked from the entry covers which models support the extended capabilities.

GitHub did not name which underlying model powers the one-million-token window or the extended reasoning mode in this announcement.

Why it matters

The 1-million-token context window is the largest single capability expansion Copilot has shipped since the AI Credits billing transition took effect June 1, 2026. For developers working in large monorepos or on projects with sprawling documentation, the practical change is that a Copilot session can now hold the full project state at once rather than requiring the developer to select a slice.

The credit cost attached to both features is the harder constraint. GitHub's billing shift, covered in detail in our June 1 report on AI Credits going live, moved all agentic Copilot work onto a per-token billing model. Choosing a larger context window means feeding more tokens to the model on every request. A one-million-token context on a large codebase could easily send several hundred thousand tokens as input on each turn, multiplying credit consumption well above a standard chat session. GitHub's recommendation to reserve the extended context and higher reasoning levels for hard problems is a practical cost-management note, not just a performance tip.

For Copilot Pro subscribers ($10/month per GitHub's billing page), the credit budget is narrower than for Pro+ ($39/month) or Max plan users. A single extended-reasoning session over a large codebase could consume a meaningful fraction of a monthly credit allocation. Until GitHub publishes per-feature credit rates, developers cannot estimate session costs before starting.

The reasoning level control solves a separate problem. Before this update, a developer asking Copilot a simple question and a developer asking it to redesign a service architecture were both routed to the same default reasoning depth. The dial gives the developer explicit control over how many credits to spend per request, which is meaningful now that every interaction has a credit price.

Context and reactions

The June 4 changelog cluster also included a separate entry covering "Fix with Copilot for failing Actions," which extends Copilot Pro, Pro+, and Max subscribers the ability to have Copilot automatically investigate a failing GitHub Actions job, push a fix branch, and tag the developer for review. Per that June 4 changelog, Copilot does this from its own cloud-based development environment without the developer staying in the workflow run page.

Taken together, the June 4 entries show GitHub shipping capability expansions within days of the billing transition rather than delaying them until developer concerns around costs settled. The pattern suggests GitHub views the AI Credits model as a platform to build on rather than a temporary billing mechanism.

The 1M-token context window also changes the competitive picture. Several competing coding assistants offer large context windows on select models, but the Copilot feature brings that capability into the VS Code extension and CLI that already sit inside most enterprise developer environments. For organizations that standardized on Copilot Business or Enterprise before the billing change, the context window expansion adds tangible value to those plans without a plan upgrade.

No public analyst comment or developer community response specific to the June 4 context window and reasoning level features was available at time of publication, given the same-day turnaround.

What to watch next

GitHub has not disclosed which models back the 1M-token context window or the extended reasoning mode. That information matters for developers comparing credit costs with equivalent capabilities on competing tools. Watch the Copilot supported models documentation for model attribution and per-model credit rates.

The first billing statements under the new AI Credits model arrive in July for monthly subscribers. Actual bills from sessions that used the extended context window or higher reasoning levels will be the first real data on how much these features cost at production usage volumes. If credit burn rates from extended context sessions prove higher than developers expect, that is likely to surface in the same community forums where the June 1 billing shift drew criticism.

Copilot subscribers can compare current plan options at /go/copilot.

Sources