COHERENCE - Science: leaderboard
Metric: Exact-match accuracy (%) on the 1,361 science documents; 6,161 interleaved image-text documents whose images were removed: the model assigns each candidate image to its placeholder; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (High) | 68.41 |
| 2 | Gemini 3.1 Pro (Preview) | 65.1 |
| 3 | Qwen 3.5 397B A17B | 60.32 |
| 4 | Claude Sonnet 4.6 (Thinking) | 58.93 |
| 5 | Kimi K2.5 | 52.09 |
| 6 | Doubao-Seed-2.0-Pro-260215 | 49.01 |
| 7 | Qwen 3.5 122B A10B | 48.79 |
| 8 | Qwen 3.5 35B A3B | 44.09 |
| 9 | GPT-5.2 | 40.78 |
| 10 | Qwen 3 VL 235B A22B | 32.77 |
| 11 | Qwen 3.5 4B | 31.45 |
| 12 | GLM-4.6V | 29.83 |
| 13 | Step3 VL 10B | 26.6 |
| 14 | Qwen 3 VL 8B Instruct | 17.56 |
| 15 | Qwen 3 VL 4B Instruct | 17.19 |
Interactive version: theaggregate.ai/benchmark?slug=coherence-science · How It Works · Data refreshed daily, snapshot 2026-10-07.