How does Copilot Fusion decide which model to use?

GitHub describes it as a runtime orchestration system that selects and coordinates models to optimize performance, cost, and latency for each task, choosing from models such as Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra, and GPT-6 Sol.

How much does Copilot Fusion cost compared with Claude Code?

In the Artificial Analysis Coding Agent Index, Copilot Fusion in GitHub Copilot costs $5.59 per task, versus $11.42 for Claude Code with Opus 5.5 (max). Token costs are charged based on the underlying models used to complete a request.

Where does Copilot Fusion rank in the benchmark?

It scores 2 points below Opus 5.5 in Claude Code in the Coding Agent Index. It leads Terminal-Bench 4.0 with a score of 71 and ranks third on FrontierCode at 51, behind Claude Code with Opus 5.5 (max) at 56 and Devin Fusion with Opus 5.5 (max) + GPT-6 Luna (high) at 54.

Large Language ModelsAI Coding AssistantsArtificial Analysis17h agoOct 29, 2026, 04:00 JST

GitHub Copilot Fusion hits $5.59 per task vs $11.42 for Opus 5.5

Send on LINE

3 Key Points

What happened
GitHub released Copilot Fusion, a runtime orchestration system that chooses among Anthropic and OpenAI models including Claude Opus 5.5 and GPT-6 Astra. In Artificial Analysis's Coding Agent Index it costs $5.59 per task, versus $11.42 for Claude Code with Opus 5.5 (max).
Why it matters
Near-frontier performance at roughly half the per-task cost suggests routing between models can cut the bill for each coding task, by inference from the $5.59 versus $11.42 figures.

WHO IT HITSDevelopers and engineering teams using AI coding agents, and the finance or platform owners who pay their per-task bills, could see lower token costs for near-frontier results if they switch to a router like Copilot Fusion.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

GitHub's Copilot Fusion is not a single new model but a runtime layer that picks among models from Anthropic and OpenAI, including Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra, and GPT-6 Sol, to balance performance, cost, and latency for each task.

Artificial Analysis benchmarked its max setting in GitHub Copilot CLI in the Coding Agent Index v2.0, where Copilot Fusion costs $5.59 per task against $11.42 for Claude Code with Opus 5.5 (max), and uses 7.2M tokens per task, about half of what Claude Opus 5.5 in Claude Code consumes. The index weights its three constituent benchmarks equally, and Copilot Fusion leads Terminal-Bench 4.0 at 71 but ranks third on FrontierCode at 51 and lags on SWE-Atlas-QnA-AA at 57.

Because token costs are charged based on the underlying models used to complete a request, the economics of the approach depend on which models the router selects for a given task, a detail GitHub's description leaves to the runtime.

FAQ

How does Copilot Fusion decide which model to use?: GitHub describes it as a runtime orchestration system that selects and coordinates models to optimize performance, cost, and latency for each task, choosing from models such as Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra, and GPT-6 Sol.
How much does Copilot Fusion cost compared with Claude Code?: In the Artificial Analysis Coding Agent Index, Copilot Fusion in GitHub Copilot costs $5.59 per task, versus $11.42 for Claude Code with Opus 5.5 (max). Token costs are charged based on the underlying models used to complete a request.
Where does Copilot Fusion rank in the benchmark?: It scores 2 points below Opus 5.5 in Claude Code in the Coding Agent Index. It leads Terminal-Bench 4.0 with a score of 71 and ranks third on FrontierCode at 51, behind Claude Code with Opus 5.5 (max) at 56 and Devin Fusion with Opus 5.5 (max) + GPT-6 Luna (high) at 54.

Artificial AnalysisRead Original Article

Send on LINE

Also reported by AI Weekly, THE BRIDGE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.