GitHub Copilot Fusion orchestrates between models at half the cost

GitHub Copilot Fusion is a runtime system that selects and coordinates Anthropic and OpenAI models for each coding task, aiming to balance performance, cost, and speed. Its available models include Claude Opus 5.5 and Sonnet 5.5, plus GPT-6 Astra and Sol. Artificial Analysis tested Fusion’s maximum setting in the Copilot CLI using its Coding Agent Index v2.0, which gives equal weight to three benchmarks and calculates task costs from the underlying models used. Fusion’s overall result was reported as two points below Claude Code running Opus 5.5 at maximum.

Fusion’s strongest showing was on Terminal-Bench 4.0, where it scored 71 and led the compared systems. It scored 51 on FrontierCode, behind Claude Code with Opus 5.5 at 56 and Devin Fusion at 54, and 57 on SWE-Atlas-QnA-AA. Its cost was $5.59 per task, compared with $11.42 for Claude Code with Opus 5.5; Devin Fusion scored 62 at $8.00 per task. Fusion used 7.2 million tokens per task - about half as many as Claude Code with Opus 5.5 and 30% fewer than Devin Fusion. The results position Fusion as a lower-cost option with near-frontier performance, rather than the top performer across every benchmark.