OMIBench - Physics: 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 physics problems; higher is better.

Metric: GPTScore (%, binarized LLM-judge equivalence of the final answer with the gold answer, micro-averaged) on the 424 physics 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
1Qwen 3 VL 235B A22B (Thinking)43.87#459
2GPT-5 Mini43.63#364
3Qwen 3 VL 32B (Thinking)41.98#573
4GPT-540.8#178
5Gemini 3 Pro (Preview)38.92#119
6Qwen 3 VL 30B A3B (Thinking)37.5#719
7O4 Mini35.38#368
8Gemini 2.5 Pro31.84#276
9Qwen 3 VL 8B (Thinking)30.66#910
10Qwen 3 VL 32B Instruct25#606
11Qwen 3 VL 235B A22B Instruct23.58#532
12InternVL3.5-38B23.58#1021
13Gemini 2.5 Flash23.35#463
14InternVL3.5-241B-A28B21.23#733
15Qwen 3 VL 30B A3B Instruct20.99#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-physics · How It Works · Data refreshed daily, snapshot 2026-11-21.