OMIBench: leaderboard
Metric: GPTScore (%, binarized LLM-judge equivalence of the final answer with the gold answer, micro-averaged) on the 1,322 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 January 2027. 37 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 50.53 |
| 2 | Qwen 3 VL 32B (Thinking) | 49.54 |
| 3 | GPT-5 | 48.11 |
| 4 | GPT-5 Mini | 47.73 |
| 5 | Qwen 3 VL 30B A3B (Thinking) | 47.2 |
| 6 | Qwen 3 VL 235B A22B (Thinking) | 47.05 |
| 7 | O4 Mini | 45.97 |
| 8 | Gemini 2.5 Pro | 44.1 |
| 9 | Qwen 3 VL 8B (Thinking) | 38.65 |
| 10 | Qwen 3 VL 32B Instruct | 35.78 |
| 11 | Gemini 2.5 Flash | 35.25 |
| 12 | Qwen 3 VL 235B A22B Instruct | 34.39 |
| 13 | Qwen 3 VL 4B (Thinking) | 30.03 |
| 14 | Qwen 3 VL 30B A3B Instruct | 28.59 |
| 15 | Qwen 2.5 VL 72B Instruct | 27.99 |
Interactive version: theaggregate.ai/benchmark?slug=omibench · How It Works · Data refreshed daily, snapshot 2026-10-07.