OMIBench - Physics: leaderboard
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.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 235B A22B (Thinking) | 43.87 |
| 2 | GPT-5 Mini | 43.63 |
| 3 | Qwen 3 VL 32B (Thinking) | 41.98 |
| 4 | GPT-5 | 40.8 |
| 5 | Gemini 3 Pro (Preview) | 38.92 |
| 6 | Qwen 3 VL 30B A3B (Thinking) | 37.5 |
| 7 | O4 Mini | 35.38 |
| 8 | Gemini 2.5 Pro | 31.84 |
| 9 | Qwen 3 VL 8B (Thinking) | 30.66 |
| 10 | Qwen 3 VL 32B Instruct | 25 |
| 11 | Qwen 3 VL 235B A22B Instruct | 23.58 |
| 12 | Gemini 2.5 Flash | 23.35 |
| 13 | Qwen 3 VL 30B A3B Instruct | 20.99 |
| 14 | Qwen 3 VL 4B (Thinking) | 20.75 |
| 15 | Qwen 2.5 VL 32B Instruct | 19.81 |
Interactive version: theaggregate.ai/benchmark?slug=omibench-physics · How It Works · Data refreshed daily, snapshot 2026-10-07.