OMIBench - Mathematics: 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 mathematics problems; higher is better.

Metric: GPTScore (%, binarized LLM-judge equivalence of the final answer with the gold answer, micro-averaged) on the 430 mathematics 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)62.56#119
2Qwen 3 VL 32B (Thinking)60.93#573
3Qwen 3 VL 30B A3B (Thinking)59.3#719
4GPT-5 Mini56.74#364
5O4 Mini56.61#368
6GPT-556.51#178
7Gemini 2.5 Pro53.49#276
8Qwen 3 VL 8B (Thinking)49.3#910
9Qwen 3 VL 235B A22B (Thinking)48.84#459
10Qwen 3 VL 4B (Thinking)44.65#1334
11Qwen 3 VL 32B Instruct40.7#606
12Gemini 2.5 Flash38.37#463
13Qwen 3 VL 235B A22B Instruct34.19#532
14Qwen 3 VL 30B A3B Instruct32.33#895
15InternVL3.5-8B32.33#1604

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

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