Atlas

Benchmarks

← All benchmarks

TxBench Antibody Discovery V1 (100 evaluations)

Science · 2026-10-09

The October 9, 2026 published V1 antibody-discovery cohort contains 100 evaluation records, reduced from the previously published 137-record cohort. Macro-mean pass rates are kept separate from that earlier cohort.

Top models (higher is better)

ModelScore
Claude Sonnet 5.557.3
GPT-6 Astra56.0
Grok 4.653.0
Grok 4.752.0
Gemini 3.7 Flash50.7
GPT-6.1 Sol49.7
Gemini 3.8 Flash49.0
Claude Opus 548.7
GPT-6 Sol43.3
DeepSeek V4.1 Flash39.7
GPT-5.6 Sol37.3
Opus 4.736.0
Opus 4.835.3
Kimi K335.3
GPT-6 Luna34.0
GPT-5.533.0
Sonnet 532.0
Sonnet 4.626.3
GPT-5.6 Luna19.3
Nemotron 3.5 Lightning 30B A3B (precision unspecified)16.0
Claude Opus 5.58.3
Loading Atlas data…