GitHub stops making you pick the right model
Until October 2026, GitHub Copilot users had to decide, task by task, which AI model to call: a fast cheap one for simple work, a slower pricier one for hard problems. At Universe 2026, the company's annual conference, GitHub confirmed that choice is no longer manual. Copilot Fusion, a system that decides on its own which model to use and how to combine more than one, moved out of preview into general availability across all paid Copilot plans, in VS Code, the Copilot app, the CLI, and JetBrains IDEs.
How Fusion decides for you
Instead of picking a specific model in the picker, developers choose an effort level: Standard for everyday work, High for tasks that need deeper reasoning, Max for the hardest problems. From there Fusion runs one of three execution modes. In Single, one model solves the task directly. In Cascade, a cheaper model drafts first and a quality check accepts the answer or escalates to a stronger model. In Critique, one model drafts, a second reviews, and the first revises. GitHub reported that Fusion Max scored 60 on Artificial Analysis's Coding Agent Index, within two points of Claude Code running Opus 5.5 Max, at roughly half the API cost, half the tokens, and 40% less time per task. That is the company's own benchmark, so it reads as a direction signal, not an independent verdict.
A new app to coordinate work, not just code
Alongside Fusion, GitHub launched a dedicated Copilot app built to organize software delivery done by AI agents, not just generate code snippets. Inside it, Copilot Projects entered public preview: users describe a goal, attach context such as repositories, design mockups, or Slack threads, pick models and tools, and get back results ready for review, including pull requests, while the agent works in parallel across more than one repository. The app also gained a computer use feature, in preview on macOS and Windows, that lets Copilot read a screen, click, and type inside apps that have no API or built in integration, always asking for approval before acting.
Deeper code review and automation that learns on its own
GitHub also introduced Copilot code review Max, aimed at code changes that need a deeper look, while keeping the lighter tiers that already existed. Automations, the cloud AI workflows triggered manually, by event, or on a schedule, got an upgrade too: they can now learn from their own outcomes and propose improvements for the team to approve, instead of just repeating the same routine every time.
Enterprise pays by credit, not by extra subscription
Starting October 28, GitHub Enterprise customers can use all of these features, including the app, code review, and Automations, without buying a separate plan: billing moves to AI credits consumed instead. Existing Copilot Business or Enterprise customers can stay on their current plan or move over whenever they want.
MaxAssistant's take
The whole package shows GitHub trying to become the platform that organizes AI agent work inside code, not just an autocomplete tool. The direct comparison to Claude Code that GitHub itself published is a company marketing its own product: useful to gauge ambition, but it calls for independent testing before any team switches tools because of it. The most concrete takeaway for developers is practical: if you already pay for Copilot, Fusion arrives at no extra cost and with a temporary 30% discount on model costs, so it is worth testing the right effort level before you keep picking models by hand.
Sources
GitHub Changelog: Route and orchestrate models with GitHub Copilot Fusion; GitHub Changelog: GitHub Copilot app, Universe releases; Unite.AI: GitHub Makes Copilot Fusion Generally Available at Universe 2026.
