GitHub Introduces Project HydraFusion for Per-Task Multi-Model Workflows in Copilot CLI

Developers enable HydraFusion by running /update, then /experimental on, then /model and selecting HydraFusion. Billing is per token consumed by whichever models the workflow invokes, at each model's standard rate. The preview has no open weights and no self-hosted path, but is available to users on all GitHub Copilot plans.

HydraFusion treats workflow selection as an optimization problem. It reads capability signals for reasoning, code generation, debugging, and tool use, then chooses the least complex workflow expected to clear a quality bar. It currently supports three execution patterns. In the single pattern, one selected model solves the task directly. In the cascade pattern, an efficient model drafts a solution, then a quality gate either accepts it or escalates to a stronger model. In the critique pattern, one model drafts, an independent read-only critic from a different model family reviews the draft, and the drafting model revises once.

GitHub built the runtime around five operating principles for repository-level work: complete accounting across every leg including drafting, critique, revision, escalation, retry, and fallback; bounded execution with explicit timeout and cancellation per leg; isolated review where critics run in tool-less contexts and cannot modify the repository; fail-safe application where no patch is applied when a workflow is cancelled or fails validation; and validated routing that checks model bindings, fallback behavior, and availability before execution starts.

According to benchmarks reported by MarkTechPost, GitHub evaluated fixed HydraFusion policies on three agentic coding benchmarks, using Claude Opus 5 and GPT-5.6 Sol as baselines at a medium reasoning level. Relative to Opus 5, HydraFusion on TerminalBench 2.1 showed an estimated cost 67% lower and a verified task quality gain of 4.9 points. On DeepSWE, the cost was 36% lower with quality down 1.5 points, while on CheckpointBench, GitHub's internal multi-turn set, cost was 65% lower with quality down 0.1 points.