GitHub has updated Copilot Fusion with three new model options, broadened the UI surface where the feature can be chosen, and published benchmark results on the public Terminal‑Bench 4.1 leaderboard. Engineers who rely on AI‑driven code assistance should note the expanded model catalog, the ability to enable Fusion in additional client environments, and the reported cost‑performance advantage over comparable frontier models.
New model lineup
The release adds the following models to the Fusion portfolio:
Claude Sonnet 5.5– available under the Fusion Max tierGPT-6.1 Terra– available under the Fusion High tierMAI-Code-2– available under the Fusion Standard tier
Each model is selected automatically by Fusion based on the effort level you set, allowing the service to route tasks to the most appropriate engine without manual prompt engineering.
Expanded surface availability
Fusion can now be activated on the latest versions of github.com, Visual Studio, and the GitHub Mobile beta apps for iOS and Android. Production mobile apps will receive support shortly. The change is delivered through a model picker UI, meaning existing CI/CD or local development workflows that already invoke Copilot can switch to Fusion with a UI selection rather than code changes.
Benchmark results and cost implications
On the public Terminal‑Bench 4.1 leaderboard, Copilot Fusion Max achieved a 71.2% task‑resolution rate, which is 5.4 percentage points higher than the leading frontier models Opus 5.5 and Fable 5.2. The same benchmark reports that Fusion Max consumed 88% less cost than Fable 5.2 and 57% less than Opus 5.5 for comparable tasks. These figures suggest a measurable efficiency gain when using Fusion for terminal‑based automation or scripting workloads.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Practitioners should evaluate whether the new model options align with their latency, token‑budget, or domain‑specific needs, and consider updating CI pipelines to select Fusion where cost‑performance is a priority. Monitoring the Terminal‑Bench leaderboard can provide an external reference point for model selection, but teams should also benchmark Fusion against internal workloads. Finally, keep an eye on the rollout of production mobile support, as that will affect on‑the‑go development and incident response scenarios.


