GitHub Copilot launches Project HydraFusion, multi-model orchestration for quality and cost

On September 4, GitHub published a blog post announcing Project HydraFusion, opened as a research preview in GitHub Copilot. HydraFusion is a runtime multi-model orchestration system that builds a full execution plan for each request, choosing models from multiple providers to draft, critique, and revise, or cascading to more powerful models to complete coding tasks.

The system treats workflow selection as an optimization problem, using capability signals for reasoning, code generation, debugging, and tool use to choose the most efficient execution pattern. For each request, HydraFusion currently picks from three execution patterns: Single, where one model solves the task directly, preserving speed and efficiency; Cascade, where an efficient model drafts a solution and a quality gate decides whether to accept it or escalate to a stronger model; Critique, where one model drafts, an independent read-only critic from a different model family reviews (following the Rubber Duck review pattern), and the drafting model revises once.

In offline evaluations across three agentic coding benchmarks, HydraFusion delivered frontier-level quality with substantial estimated cost savings: on TerminalBench 2.1, verified task quality improved by 4.9 percentage points at 67% lower estimated cost than Claude Opus 5; on DeepSWE, estimated cost dropped 36% with quality down 1.5 points; on the internal CheckpointBench, estimated cost dropped 65% with quality down 0.1 points.

To use it, developers run /update in GitHub Copilot CLI to install the latest version, /experimental to enable, then /model and select HydraFusion (Research Preview). HydraFusion remains a research preview; GitHub notes that usage is billed at the standard per-token rate of each model HydraFusion uses, and welcomes community feedback to iterate on the model pool and workflow configuration.