Anthropic protein design verified: Opus 4.8 hits 35% rate via tool use, Chai-2 still leads
eyishazyer · x · 2026-08-19
This post clarifies and analyzes Anthropic's recent protein design results.
Key Facts:
- Model Correction: The work was done using Mythos Preview and Opus 4.8, not Opus 5.
- Workflow: Claude did not design structures directly but drove existing open-source design and folding tools, performing at an expert level.
- Benchmark: The 22-35% hit rate (independently verified) is an improvement over the standard 10-15%, but still trails Chai Discovery's Chai-2, which achieved 68% on the same task type over a year ago.
Limitations:
- Zero of 90 designs bound to maltose-binding protein.
- Anthropic openly admitted they don't know why Opus 4.8 succeeded where Mythos Preview failed, a rare display of honesty.
Related event: Anthropic Says Claude Designed Protein Binders for 14 of 15 Targets(45 posts)→
More from Models
- Nolla Health Post-Trains Nemotron 3.5 VL: Visual Diagnosis Jumps from 35% to 66.8% — LoganGrasby · 2026-10-16
- Dev argues coding models now beat nearly all engineers in every branch of software — zack_overflow · 2026-10-16
- Prediction: Chinese frontier models could match GPT-6 by March 2027 — teortaxesTex · 2026-10-16
- Liquid AI's d1-ultra, #1 on Artificial Analysis Decision Index, lands on Vercel — JosephJacks_ · 2026-10-16
- Weekly AI roundup: Claude Haiku 5.5 cuts prices ~90%, Gemini 4 Argon output jumps to 1M tokens — DeepLearningAI · 2026-10-16
- Tencent Hunyuan says reported agent activity came from third-party model evals, not deliberate targeting — TencentHunyuan · 2026-10-16