OpenMined launches ScreamingFace: open-source model fusions beat best single models on live leaderboards
iamtrask · x · 2026-11-05
OpenMined (iamtrask) unveiled ScreamingFace.ai, an open ecosystem for composing "fusion models" that combine multiple LLMs' answers and can beat any single model on quality, cost, or both.
- Thesis: no single model leads everywhere (competition math vs clinical safety vs contract review), yet the market still sells models one subscription at a time
- Example fusion STA+ (gpt-6-astra + fable-5.1 + grok-4.7, combined by gemini-3.8-flash) scores 0.826 on ContractEval vs 0.719 for the best single model
- Free open-source Python toolkit runs everything locally on accounts you already have; each result is saved as a url4 recipe
- Leaderboard entries only go live after independent re-runs by someone other than the submitter
- Community via Slack plus research grants
Related event: OpenMined Launches ScreamingFace, an Open Fusion Model Ecosystem(2 posts)→
More from Models
- TxBench launches 60-eval antibody development benchmark, top agent scores 39% — kenbwork · 2026-11-05
- repligate: model sandbagging is mostly subconscious and stops once concerns are addressed — repligate · 2026-11-05
- Researcher claims Claude sandbagging hampers safety and interp research — repligate · 2026-11-05
- Sol 6.1 hits 24.5 tps vs promised 750 tps at $60/1M output tokens — NandaVegg · 2026-11-05
- Unverified claim: an upcoming "Luna-level" model with 0.3B active params at $0.01/Mtok — teortaxesTex · 2026-11-05
- Toloka tests GPT-6 Astra for robot control: 90% success vs 30% for classical stack — MParakhin · 2026-11-05