GitHub's HydraFusion tops Claude Opus 5 by 4.9 pts on Terminal-Bench while cutting cost 67%

mariorod1 · x · 2026-09-05

GitHub launched Project HydraFusion, a research preview that delivers frontier coding quality via runtime multi-model orchestration. In controlled offline evaluations it scored 4.9 percentage points higher than the Claude Opus 5 baseline on Terminal-Bench 2.1 while cutting estimated workflow cost by 67%. It treats workflow selection as an optimization problem, using capability signals (reasoning, codegen, debugging, tool use) to pick among three patterns: Single (one model), Cascade (cheap draft with a quality gate that escalates to stronger models), and Critique (draft plus an independent read-only critic from another provider). It builds on Auto model selection toward automated semantic routing across local, cloud, and compound models.

Related event: GitHub Copilot Debuts HydraFusion Multi-Model Orchestration(3 posts)→

Original post →

More from coding & agent

coding & agent channel →