GitHub HydraFusion is a multi-model research preview that performs model orchestration inside the GitHub Copilot CLI for developers who want strong coding results at lower token cost. Launched on September 4, 2026, it picks among three execution patterns per task instead of calling one fixed model for everything.
Core Features
- Three patterns: Single runs one model; Cascade tries a cheaper model first and escalates only when quality falls short; Critique has one model write code and a second review it.
- Cost control: GitHub reports 36% to 67% lower token cost than a single Claude Opus 5 on TerminalBench 2.1, DeepSWE, and CheckpointBench.
- Benchmark parity: it matches or beats Claude Opus 5 on those three coding benchmarks while spending less.
- No new billing: it runs through standard Copilot subscription rates with no separate charge.
- CLI access: subscribers invoke it with the /experimental command in the Copilot CLI.
Use Cases
- Teams on any Copilot tier wanting cheaper, routed coding help.
- Repetitive refactors where a cascade pattern avoids overspending on easy steps.
- Shops that want a reviewer model catching issues before code ships.
Pricing
HydraFusion adds no direct fee; usage is billed at the standard rates of the underlying models under an existing GitHub Copilot plan. Free, Pro, Business, and Enterprise Copilot tiers can reach the preview through the CLI.
Our Take
Best for Copilot subscribers who want model routing without a new bill. The trade-off is that it is a research preview, so behavior and availability may change, and it lives only inside the CLI for now.
Find more in our AI Programming development and AI Engine/Model categories.
