GR-Ben - Biology: leaderboard

GR-Ben Biology is the subset of GR-Ben (2026) built from the biology problems (MMLU-Pro): a model reads a problem and a step-by-step solution written by one of 15 LLMs and must name the earliest erroneous step or state that every step is correct. The score is F1 in percent, the harmonic mean of accuracy on flawed and on correct solutions, higher is better; process reward models are scored from their step predictions and LLMs are prompted as critics with thinking disabled.

Metric: F1 (%) of first-error identification: the harmonic mean of the accuracy on flawed solutions (the earliest erroneous step named) and the accuracy on fully correct solutions (judged correct), on the 399 biology problems (MMLU-Pro) of GR-Ben (3,595 human-annotated solutions written by 15 LLMs); process reward models by their step predictions (scalar-score PRMs thresholded on a 40-item dev split held out of their scoring), LLMs prompted as critics with thinking disabled (mean of three runs for open models, one run for API models); higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 22 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 FlashMinimal57.6#180
2DeepSeek V3.2Non-reasoning49.9#425
3GPT-5.2 (2025-12-11)Non-reasoning48.9
4Kimi K2 090547.4#578
5Gemma 3 27B (IT)44.8#1341
6VersaPRM (8B)44.3#1595
7Qwen 3 32BNon-reasoning41.7#983
8Qwen 3 14BNon-reasoning38.1#1284
9Qwen 3 8BNon-reasoning37.5#1890
10Gemma 3 12B (IT)36.4#1907
11Qwen 3 4BNon-reasoning35.1#2379
12Qwen2.5-Math-PRM-72B33.6
13Llama 3.3 70B Instruct27.7#1394
14Universal-PRM-7B24.4
15Qwen2.5-Math-PRM-7B22.9

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=gr-ben-biology · How It Works · Data refreshed daily, snapshot 2026-11-22.