GitHub Copilot Fusion nears Opus 5.5 at half cost
GitHub's new Copilot Fusion routes coding tasks between Anthropic and OpenAI models and scores two points below Claude Code with Opus 5.5 while costing half as much per task.
GitHub has released Copilot Fusion in GitHub Copilot. It is a runtime orchestration system that chooses and coordinates models from Anthropic and OpenAI for each task, with the goal of balancing performance, cost and latency. Its model pool includes Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra and GPT-6 Sol. Users pay token costs for whichever underlying models handle a request.
Artificial Analysis benchmarked Fusion's max setting in GitHub Copilot CLI on its Coding Agent Index v2.0. Fusion scored 2 points below Claude Code with Opus 5.5 (max). It cost $5.59 per task, compared with $11.42 for Claude Code. Devin Fusion, which pairs Opus 5.5 (max) with GPT-6 Luna (high), scored 62 at $8.00 per task.
The index gives equal weight to three benchmarks, and Fusion's results varied across them:
- Terminal-Bench 4.0: Fusion led narrowly with 71.
- FrontierCode: Fusion placed third with 51. Claude Code with Opus 5.5 scored 56 and Devin Fusion scored 54.
- SWE-Atlas-QnA-AA: Fusion trailed with 57, which put it between GPT-6 Astra in Codex and Gemini 4 Argon in Antigravity CLI.
Lower token use explains much of the cost gap. Fusion used 7.2M tokens per task, about half of what Opus 5.5 used in Claude Code and 30% less than Devin Fusion.
Artificial Analysis tested only the max setting. Its published material does not say how Fusion decides which model to use, how often each model is selected, or how lower settings perform.