New: Claude Fable 5.2 eval report is here.Read the report ↗
Compare models

Best LLMs for computational protein design

De novo binder design

Tests computational protein-binder design within a fixed time and hardware budget, helping research teams assess AI support for generating candidates. Scores do not establish experimentally validated binding or clinical usefulness.

Updated: Data as of:
Limited reportingNot saturated
Official benchmark
De novo binder design results. Each row identifies its benchmark version and evaluation track. Scores from different versions are not directly comparable. Best effort from the latest stored report date per model and version.
Model / creatorVersionReported score
Claude Mythos 5.2Anthropic · Proprietary15 targets / 24 hours / H20085.1%
Claude Opus 5.5Anthropic · Proprietary15 targets / 24 hours / H20082.6%
Claude Mythos 5.1Anthropic · Proprietary15 targets / 24 hours / H20079.1%
View all 3 published reports
15 targets / 24 hours / H200

15 targets; 30 binders each; one H200 GPU and 24 hours per target; safeguards off. · Tools enabled

Anthropic ↗
15 targets / 24 hours / H200

effort: highest available · safeguards: false · November 9 launch-table comparison; highest available effort unless explicitly noted. Preserve the benchmark track and model configuration; competitor figures are as reported by Anthropic. · Tools enabled

Anthropic ↗
15 targets / 24 hours / H200

effort: highest available · safeguards: false · November 9 launch-table comparison; highest available effort unless explicitly noted. Preserve the benchmark track and model configuration; competitor figures are as reported by Anthropic. · Tools enabled

Anthropic ↗

Sources and original reports ↗