Opus 5 tops nine biology benchmarks, but still trails in some analysis tasks
kenbwork · x · 2026-07-25
A benchmark run on Opus 5 across nine agentic biology tasks says the model is the best evaluated so far on several areas, including variant discovery, preclinical pharmacology, genomic surveillance, and long-horizon single-cell analysis.
Key results
- Variant discovery and interpretation: +7.6 pp over Sol 5.6
- Small-molecule preclinical pharmacology: +7.0 pp over Sol
- Genomic surveillance: +7.3 pp over Sol
- Long-horizon single-cell analysis: +3.2 pp over Sol
The analysis says Opus 5 beats every previous Anthropic model on the suite, taking 6 of 7 benchmarks from Opus 4.8, often by 8–16 points. But it is not uniformly better:
- On short-horizon single-cell analysis, it trails Sol 5.6 by 2 points (60.1% vs 62.1%).
- On long-horizon spatial biology, it more than doubles Opus 4.8 (25.0% vs 11.1%) but still loses to Sol (38.9%).
- It also regresses on epigenomics analysis, where Opus 4.8 is 4.2 points higher.
The post is a continuation of a larger article on how good Opus 5 is at biology, with the benchmark dashboard now live on benchmarks.bio.
More from Models
- DeepSeek cut KV cache per token from 389KB to 890B in two years as open-closed gap debate rages — teortaxesTex · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27
- Unverified: fable 5.5 speculated at 4-5T parameters to outscale 1-3T Opus — haider1 · 2026-10-27
- Kimi K3 Is 'Open' Until You Succeed: License Caps Free Use at $20M Revenue — alex_verem · 2026-10-27
- Screenshot shows 'GPT-6 Astra Ultra' in ChatGPT model picker — unverified leak or joke — justalexoki · 2026-10-27
- DeepSeek Safety Eval Finds "Over-Attribution": Model Mistakes Ordinary Tasks for Tests — Much_Boysenberry9124 · 2026-10-27