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.
More from coding & agent
- Open models below 4-bit struggle at self-iterating 3D asset generation, HF dev finds — mervenoyann · 2026-10-21
- Developer Moves AI Agents to a Dedicated Linux Workstation Using Incus System Containers — kevinkern · 2026-10-21
- Motion Launches a General Video Agent That Edits Footage and Mimics Any Reference Style — mhdfaran · 2026-10-21
- MCP Platform NitroStack Wins Product Hunt Product of the Day, Focuses on Retention — EstablishmentFun4373 · 2026-10-21
- CodeRabbit Ships 5,000 PRs a Week With a Tiny Team, Hires by Strategy-Game Hours — omojumiller · 2026-10-21
- LukeW maps the five UI problems every AI-native company faces with agents — LukeW · 2026-10-21