Free tools Windows power users keep installed
One-click scans. No signup required.
Project HydraFusion is a research preview in GitHub Copilot CLI that decides how to handle a coding request—not just which model to call. Depending on the task, it may use one model, draft and escalate if needed, or ask a separate model to critique a draft before the original model revises it.
What HydraFusion does
GitHub senior product manager Andrea Liliana Griffiths describes HydraFusion as a workflow router for Copilot CLI. Its goal is to use the lightest workflow likely to meet a quality bar, balancing the chance of a better result against the extra model work, cost, and time that additional steps require. It is not a standalone coding editor.
The three paths in Griffiths’s September 21, 2026 explainer are Single, Cascade, and Critique. The router chooses among them for a request; the descriptions do not establish that every task follows all three paths or that users select a particular path themselves.
How the three paths differ
| Path | What happens | When the extra work may make sense |
|---|---|---|
| Single | One model attempts to solve the task. | A direct run is a reasonable fit when the request appears straightforward enough not to warrant additional review or escalation. |
| Cascade | An efficient model drafts a solution. A quality gate assesses it and may escalate the task. | Useful when a first attempt may be sufficient, but a quality check could justify a more capable follow-up. Escalation is conditional, not automatic. |
| Critique | A separate model family reviews the draft in a read-only context without tools. The original drafter then revises once. | Potentially useful when independent criticism could catch issues that another unaided attempt might miss. Review and revision add model work. |
These are workflow patterns, not a published head-to-head ranking. The source material does not provide independent consumer testing that compares quality, cost, and latency across all three paths.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →#1 Best Overall
Why use more than one model step?
A single model call is simpler, but it may not be the best choice for every task. Cascade gives an initial draft a chance to pass a quality gate before any escalation; Critique adds a separate perspective and one revision. The point is to spend extra work where it may improve the answer rather than apply the most involved process to every request.
That trade-off is real: each extra leg can mean another model call, more elapsed time, and higher cost than a single run. HydraFusion’s stated aim is not to make every task cheaper, but to route work in a way that balances quality with those added costs.
What GitHub’s benchmark does—and does not—show
GitHub reported that, on TerminalBench 2.1, HydraFusion improved verified task quality by 4.9 percentage points at 67% lower estimated cost compared with Claude Opus 5. These are reported offline-evaluation results against that named baseline, not a promise about the cost or quality of an individual Copilot CLI request. GitHub’s release statement is available in an indexed reproduction of its September 4, 2026 release; the underlying GitHub page was not independently accessible in the reviewed material.
In particular, “lower cost than always running Opus” does not mean HydraFusion will cost less than choosing one inexpensive Auto option for a small task. Cascade and Critique may add calls, and the result depends on the task and routing. Griffiths also says she is still testing token use against manually passing context among models, so the benchmark should not be read as a general comparison with every alternative workflow.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsSafeguards described for the preview
Griffiths’s explainer lists runtime safeguards intended to manage the additional steps. These are descriptions of the system’s stated behavior, not independently verified test results:
- Cost accounting covers every leg of a workflow.
- Timeouts and cancellation are supported.
- Critique runs in an isolated, tool-less review context.
- A failure or cancellation does not produce a patch.
- Routing is validated before execution.
Who should try it first
Griffiths recommends starting with a well-scoped coding task on the first turn in Copilot autopilot. That gives the router a bounded request to handle without making multi-turn polishing the test case. The explainer describes multi-turn polishing as future work; it does not establish how well HydraFusion handles that workflow.
Rank #4
HydraFusion is a research preview, and its names, model pool, and behavior may change. For current availability and operating details, check GitHub’s Copilot CLI information. The preview explainer also points users to /feedback in Copilot CLI and to a GitHub Community discussion.
Quick Recap
Best Value
Sources
- Andrea Liliana Griffiths, “Project HydraFusion, in plain English,” DEV Community, September 21, 2026.
- GitHub, “Project HydraFusion: Frontier quality via multi-model orchestration,” September 4, 2026 (indexed reproduction).
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
