Claude Opus 5 claims strong benchmark gains across coding, search, and reasoning
natolambert · x · 2026-07-25
Anthropic’s Claude Opus 5 is presented as a stronger frontier model, with the post citing faster iteration and scaled RL as key contributors.
The attached benchmark chart shows Opus 5 leading or tying across several tasks, including:
- Agentic terminal coding: 43.3%
- GDPVal-AA v2: 1861
- ARC-AGI-3: 30.2%
- BrowseComp: 90.8%
- Humanity’s Last Exam: 56.3% without tools, 64.7% with tools
- OSWorld 2.0: 70.6%
- FrontierCode v1.1 Main: 53.4%
- AutomationBench: 26.0%
- Legal Agent Benchmark: 11.7%
- HealthBench Professional: 59.8%
- BioMysteryBench: 49.4% hard, 90.1% human solved
The post also claims safeguards classifiers should intervene about 85% less often than they do for Fable 5.
Related event: Anthropic Releases Claude Opus 5: SOTA Performance at Half the Price(128 posts)→
More from Models
- MissionReport benchmark: only 3 of 540 AI-generated government reports pass all criteria — Exp_Mark · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27
- Nat Lambert: Cohere's model is narrow RL training on internal infra ill-suited for open weights — natolambert · 2026-10-27
- OpenAI teases DOTS is coming to Europe 'soon' — mark_k · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27