GitHub has launched Project HydraFusion, a research preview for GitHub Copilot CLI that introduces multi-model orchestration. Instead of routing prompts to a single model, the system dynamically builds a workflow per coding task using models from multiple providers. It operates on three patterns—Single, Cascade, and Critique—to balance quality and cost. Benchmarks show HydraFusion improves quality by 4.9 points on TerminalBench 2.1 while cutting costs by up to 67% compared to Claude Opus 5.
Project HydraFusion multi-model orchestration
- ▪GitHub released Project HydraFusion, a research preview that builds a multi-model execution plan per request instead of routing prompts to a single model.
- ▪Project HydraFusion reads capability signals for reasoning, code generation, debugging, and tool use to select the least complex workflow expected to clear a quality bar.
Three execution patterns
- ▪Project HydraFusion selects from three execution patterns: Single, where one model solves the task; Cascade, which drafts and escalates; and Critique, which drafts and reviews.
- ▪Under the Critique execution pattern in Project HydraFusion, an independent read-only critic from a different model family reviews a draft and the drafting model revises it once.
Runtime engineering guardrails
- ▪The Project HydraFusion runtime logs role, outcome, cost, latency, and diagnostics internally per leg, while presenting the developer with one coherent response and permission-aware change set.
- ▪GitHub built the Project HydraFusion runtime around five operating principles: complete accounting, bounded execution, isolated review, fail-safe application, and validated routing.
Benchmark performance results
- ▪On the DeepSWE and CheckpointBench benchmarks, Project HydraFusion trailed Claude Opus 5 slightly while reducing costs by 36% and 65% respectively.
- ▪On the TerminalBench 2.1 benchmark, Project HydraFusion achieved a 4.9 quality point increase at a 67% lower estimated cost compared to Claude Opus 5.
Copilot CLI availability
- ▪Project HydraFusion is available as a research preview for users on all GitHub Copilot plans exclusively inside the GitHub Copilot CLI.
- ▪To activate Project HydraFusion in the GitHub Copilot CLI, users must run the `/update`, `/experimental on`, and `/model` commands.
Per-token billing model
- ▪Billing for Project HydraFusion is calculated per token consumed by whichever models the workflow invokes, charged at each model's standard rate.
- ▪Project HydraFusion invokes models from multiple providers, and users are billed based on the standard rates of the specific models utilized in the execution plan.
Story comments
Loading comments…