Anthropic's Agent4Sci: Opus Outperforms Published SOTA on 20 Scientific ML Benchmarks

Anthropic and its Fellows Program released Agent4Sci, a long-horizon benchmark spanning 20 scientific ML tasks across 11 domains, where the Opus coding agent beat all published state-of-the-art results. The evaluation notes the gains came mainly from engineering tricks rather than genuine domain insights.

2026-10-22 ~ 2026-10-22 · 2 related posts

• Anthropic's Agent4Sci: Frontier Agents Beat SOTA on 20 Scientific ML Benchmarks, but Only via Engineering Tricks — Anthropic · 2026-10-22
• Anthropic's Agent4Sci: Claude Opus beats published SOTA on all 20 science ML benchmarks — dair_ai · 2026-10-22