BenchBench-Protocol - Protein Biochemistry and Purification: leaderboard

Metric: Normalized rubric score (%; weighted rubric elements met by the response to real protocol-modification tasks in the protein biochemistry and purification domain (32 tasks), graded by GPT-5.6 Terra; mean of repeated attempts per task). Source: arxiv.org. Saturation forecast: Around April 2028. 9 models tracked.

Top models

#ModelScore
1Claude Opus 5 (Max)58.1
2GPT-5.6 Sol (Max)43
3Kimi K3 (Max)41.6
4GPT-5.6 Terra (Max)41.3
5GPT-5.6 Luna (Max)39.4
6Inkling (Max)37.4
7GLM-5.2 (Max)37.2
8Grok 4.5 (High)36.3
9Gemini 3.6 Flash (High)31.8

Interactive version: theaggregate.ai/benchmark?slug=benchbench-protocol-protein-biochemistry-and-purification · How It Works · Data refreshed daily, snapshot 2026-09-26.