BenchBench-Protocol - Cell Culture and Cell-Based Assays: leaderboard

Metric: Normalized rubric score (%; weighted rubric elements met by the response to real protocol-modification tasks in the cell culture and cell-based assays domain (27 tasks), graded by GPT-5.6 Terra; mean of repeated attempts per task). Source: arxiv.org. Saturation forecast: Around January 2028. 9 models tracked.

Top models

#ModelScore
1Claude Opus 5 (Max)58.4
2Kimi K3 (Max)46.7
3GPT-5.6 Terra (Max)46.6
4GPT-5.6 Sol (Max)46.5
5GPT-5.6 Luna (Max)38.7
6GLM-5.2 (Max)36.8
7Inkling (Max)36.7
8Grok 4.5 (High)36.5
9Gemini 3.6 Flash (High)32.8

Interactive version: theaggregate.ai/benchmark?slug=benchbench-protocol-cell-culture-and-cell-based-assays · How It Works · Data refreshed daily, snapshot 2026-09-26.