Copilot Fusion Hits Near-Opus Coding Agent Scores at About Half the Cost per Task. You Do Not Pick the Model, and Discarded Drafts Can Leave File Edits Behind.
GitHub Copilot Fusion went generally available on October 28 in VS Code, the Copilot CLI, the Copilot app, and JetBrains: a runtime router across Claude Opus and Sonnet 5.5, GPT-6 Astra and Sol, and others, with Single, Cascade, and Critique patterns and Standard, High, and Max effort. Artificial Analysis scores Fusion Max at 60 on its Coding Agent Index, within two points of Claude Code with Opus 5.5, at $5.59 per task against $11.42. GitHub's docs say a discarded draft does not undo file edits already made, the context window is the smallest member model's, and paid plans get 30% off Fusion model costs through November 30. We read the changelog, the docs, and AA's article.
By T.W. Ghost
GitHub Copilot Fusion is generally available today, October 28, 2026, in VS Code, the Copilot CLI, the Copilot app, and JetBrains. Fusion is not a model. It is a runtime router that picks across Claude Opus 5.5 and Sonnet 5.5, GPT-6 Astra and Sol, and others, runs them in one of three patterns, Single, Cascade, or Critique, at one of three effort levels, Standard, High, or Max, and bills you for whichever underlying models it used at their standard rates.
GitHub's changelog leads with a number from Artificial Analysis: Fusion Max scored 60 on AA's Coding Agent Index, within two points of Claude Code with Opus 5.5 at max, at about half the average cost. AA's own article puts the cost at $5.59 per task against $11.42 for Claude Code with Opus, with Fusion using about 7.2 million tokens per task, roughly half of Opus in Claude Code. The catches are in GitHub's documentation, not the changelog: you do not choose the model; the context window you get is the smallest member model's; and if Fusion discards a draft, the file edits that draft already made are not undone. We read the changelog, the Fusion docs, the local-models changelog, and AA's article. Here is what held up.
What You Are Turning On
| Copilot Fusion (GA October 28, 2026) | |
|---|---|
| Where | VS Code, Copilot CLI, Copilot app, JetBrains |
| Models in the pool | Claude Opus 5.5, Sonnet 5.5, GPT-6 Astra, GPT-6 Sol, and others; local MAI Code 1.1 Flash or Qwen3.8 Flash Next can enter Standard when installed |
| Patterns | Single, Cascade, Critique |
| Effort | Standard, High, Max; High and Max can use more AI credits |
| Billing | Per underlying model at standard rates; 30% off Fusion model costs on paid plans through November 30, 2026 |
| Context window | The smallest member model's |
| Discarded drafts | File edits already made are not undone (docs) |
| Model choice | None; the router decides, and the routing changes over time |
The discard warning is the one to read twice. Cascade and Critique generate drafts and throw some away. A thrown-away draft that already wrote to disk leaves those writes behind. Review the working tree before you commit, not the chat transcript.
Smallest-member context is a hidden cap. Your effective context is not Opus's or Astra's; it is whatever the least capable model in the pool can hold.
The promo is a cost line with an end date. Thirty percent off through November 30. AA's $5.59 may or may not already net it out; AA's article does not say, and we do not assume.
The Benchmarks, Sorted by Who Reported Them
| Claim | Number | Who |
|---|---|---|
| Coding Agent Index, Fusion Max | 60 | Independent (AA; GitHub cites it) |
| Claude Code with Opus 5.5 (max) | about 62 | Independent (AA) |
| Cost per task, Fusion Max vs Claude Code with Opus | $5.59 vs $11.42 | Independent (AA) |
| Tokens per task | 7.2M, about half of Opus in Claude Code | Independent (AA) |
| Terminal-Bench 4.0 | 71, leads the board | Independent (AA) |
| FrontierCode | 51 | Independent (AA; private Cognition benchmark) |
| SWE-Atlas-QnA-AA | 57 | Independent (AA) |
| Devin Fusion | 62 at $8.00 per task | Independent (AA) |
Three honest observations:
- AA tested Max. Standard and High route differently and will score differently, and most paid seats will run Standard. The 60 is the ceiling configuration.
- Near-Opus at half the cost is the headline, and both halves are AA's. That is the rare case this month where the vendor's number and the independent number are the same number.
- Devin Fusion is the comparison GitHub does not make. On the same AA article, Cognition's router scores 62 at $8.00 a task: two points higher, $2.41 more. Cognition also builds FrontierCode, one of the rows both routers are scored on.
Verified vs Unconfirmed: The Scorecard
| Claim | Verdict |
|---|---|
| GA Oct 28 across VS Code, CLI, app, JetBrains; patterns and effort levels | Verified (GitHub changelog and docs) |
| Fusion Max 60 on Coding Agent Index; about half the cost of Opus Max | Verified (GitHub changelog citing AA; AA article) |
| $5.59 vs $11.42; Terminal-Bench 4.0 71; FrontierCode 51; SWE-Atlas 57; 7.2M tokens per task | Verified (AA article) |
| Devin Fusion 62 at $8.00 | Verified (AA article) |
| 30% off paid plans through Nov 30, 2026 | Verified (changelog and docs) |
| Discarded drafts do not undo file edits; smallest-member context | Verified (GitHub docs, limitations) |
| Local MAI Code 1.1 Flash and Qwen3.8 Flash Next in Standard | Verified (docs and local-models changelog) |
| Whether $5.59 nets out the 30% promo | Unverified |
| Plan availability matrix | Unverified (blank in some fetches) |
Who Should Use It
Use Fusion if you already pay for Copilot and want near-Opus agent results at a lower per-task cost without choosing models yourself; run Max for the benchmarked behavior and read the diff before every commit.
Skip it if your work needs a specific model, a specific context window, or a guarantee that a rejected attempt left nothing on disk. And re-price it on December 1.
The frontier is elsewhere: the single best agent on AA's board is still Claude Code with Opus 5.5 at 62, and the router that comes closest for less is the one you do not control. Our model comparison has the current standings side by side.
Sources
- GitHub changelog: route and orchestrate models with GitHub Copilot Fusion
- GitHub docs: Copilot Fusion
- Artificial Analysis: GitHub Copilot Fusion orchestrates Anthropic and OpenAI models
- GitHub changelog: Copilot Fusion now supports selection of local models
*Not sure whether a router or a single model fits how your team ships? Take the free 2-minute quiz and get matched. Then read about the index Fusion is scored on and the private benchmark inside it: AA's Coding Agent Index v2.0.*