OMIBench - Biology: leaderboard

OMIBench (2026) tests vision-language models on Olympiad-level science and mathematics problems whose evidence is spread across several images. This board is the GPTScore in percent on the biology problems; higher is better.

Metric: GPTScore (%, binarized LLM-judge equivalence of the final answer with the gold answer, micro-averaged) on the 251 biology problems of OMIBench (Olympiad-level multi-image problems from biology, chemistry, mathematics and physics competitions), zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 37 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro (Preview)71.31#119
2Gemini 2.5 Pro66.14#276
3Gemini 2.5 Flash64.54#463
4Qwen 3 VL 32B (Thinking)64.14#573
5Qwen 3 VL 235B A22B Instruct63.2#532
6GPT-562.55#178
7Qwen 3 VL 235B A22B (Thinking)61.35#459
8GPT-5 Mini59.36#364
9Qwen 3 VL 32B Instruct58.57#606
10O4 Mini57.37#368
11Qwen 3 VL 30B A3B (Thinking)55.78#719
12Qwen 2.5 VL 72B Instruct53.39#841
13GPT-4o53#751
14Qwen 3 VL 8B (Thinking)52.59#910
15Qwen 3 VL 30B A3B Instruct48.61#895

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=omibench-biology · How It Works · Data refreshed daily, snapshot 2026-11-21.