From the source
GitHub is expanding Copilot Fusion to more Copilot surfaces and adding three models to its tiers, while publishing results on two public Terminal-Bench leaderboards.
- Copilot Fusion is now listed on Terminal-Bench 4.1 leaderboards hosted by Stanford, Harbor and the Laude Institute, which measure how well AI agents complete hard, realistic terminal tasks.
- GitHub says Copilot Fusion Max reached a 71.2% resolution rate there, 5.4 percentage points ahead of leading frontier models Opus 5.5 and Fable 5.2, while costing 88% less than Fable 5.2 and 57% less than Opus 5.5.
- The company says independent leaderboards let users compare Copilot with other agents on the same tasks under the same evaluation rules.
- Users can now select Copilot Fusion on the latest versions of github.com, Visual Studio, and the GitHub Mobile beta apps for iOS and Android, with support in the production GitHub Mobile apps coming shortly.
- In the model picker, Copilot chooses and coordinates models for each task while the user sets the effort level.
- The release adds Claude Sonnet 5.5 to Copilot Fusion Max, GPT-6.1 Terra to Copilot Fusion High, and MAI-Code-2 to Copilot Fusion Standard.





