October 2, 2026
Copilot connects models: HydraFusion checks and refines answers
HydraFusion can assign a draft to one model and a review to another, instead of having a single model answer.
GitHub
@github
Project HydraFusion and new models are now in the Copilot model picker. Have you tried them yet? 👀 https://x.com/i/article/2106059231531319298
· 22.2K views
On Sep 30, GitHub added HydraFusion to GitHub Copilot for VS Code and Copilot app. Previously, it ran in Copilot CLI.
Models work together. In Auto mode, Copilot selects one model per request. HydraFusion chooses one of three ways to handle a task:
- Single: one model prepares the answer. - Cascade: a model prepares a draft, then HydraFusion coordinates a review and hands the task to a stronger model if needed. - Critique: one model prepares a draft, another critiques it, followed by one revision.
In GitHub's controlled offline test on Sep 4 using TerminalBench 2.1, HydraFusion improved task completion performance by 4.9 percentage points over Claude Opus 5. Estimated task execution costs fell by 67%.
How to enable HydraFusion
HydraFusion is available in preview on Pro, Pro+, Business and Enterprise plans. For Business and Enterprise, an administrator must enable preview features.
- In VS Code 1.140+ or Insiders, enable `chat.copilot.hydraFusion.enabled` in Settings, then select HydraFusion from the Copilot Chat model menu. - In GitHub Copilot app, update to the latest version, enable HydraFusion in Settings → Experimental and select it from the model menu. - In Copilot CLI, run `copilot update`, launch `copilot --experimental` and use `/model` to select HydraFusion labeled Research Preview.
For a single request in CLI, use `copilot --experimental --model hydrafusion -p "YOUR-PROMPT"`. `/usage` shows usage for the models involved.
GitHub also added GPT-6.1 Sol and Claude Sonnet 5.5 to the model menu. Both are available in VS Code, Copilot CLI and Copilot app, with access rolling out gradually.
- GPT-6.1 Sol: from Sep 29 on Pro+, Max, Business and Enterprise. - Claude Sonnet 5.5: from Sep 28 on Pro, Pro+, Max, Business and Enterprise.
How much model usage costs
Each model is billed separately. HydraFusion is billed at the standard rates for all models involved. There is no separate orchestrator fee, and the Auto discount does not apply.
GitHub's rates as of Sep 30, 2026:
| Model and input size | Input per 1 million tokens | Output per 1 million tokens | | --- | --- | --- | | GPT-6.1 Sol, up to and including 272,000 tokens | $2 | $10 | | GPT-6.1 Sol, over 272,000 tokens | $4 | $15 | | Claude Sonnet 5.5 | $2 | $10 |
GitHub bills caching separately. One AI credit equals $0.01.
On Business and Enterprise, an administrator controls access to new models through the model policy in Copilot settings.
Source
