GitHub Copilot Fusion adds three models and its first benchmark
GitHub adds Claude Sonnet 5.5, GPT-6.1 Terra and MAI-Code-2 to its Copilot model router, and publishes a Terminal-Bench score for the first time.
GitHub added three models to Copilot Fusion, the model router it took out of preview on 28 October. Claude Sonnet 5.5 joins the Fusion Max tier, GPT-6.1 Terra joins Fusion High, and MAI-Code-2 joins Fusion Standard, giving each effort level a newer option to route to.
Fusion is also now available on github.com, in Visual Studio, and in the GitHub Mobile beta apps for iOS and Android, with full mobile support due shortly. GitHub published its first public benchmark for the router: on Terminal-Bench 4.1, Fusion Max resolved 71.2% of tasks, 5.4 points ahead of Opus 5.5 and Fable 5.2, while costing 88% less than Fable 5.2 and 57% less than Opus 5.5.
Fusion Max -> Claude Sonnet 5.5
Fusion High -> GPT-6.1 Terra
Fusion Standard -> MAI-Code-2
Why it matters: Fusion exists so you don’t have to pick a model by hand for every coding task. A published benchmark with cost figures attached gives you something concrete to weigh against just defaulting to a named flagship model, especially for terminal-heavy agentic work where the cost gap compounds fast.
The caveat: the benchmark is GitHub’s own, run on its own router, against models it has a commercial reason to look favourable against.