MuseBench - Single-Select (Overall): leaderboard
MuseBench evaluates intent-level audiovisual-arts understanding in multimodal models over 4,016 questions from video essays in four art categories (cinematic, static visual, stage performing, game arts), with single-select and multi-select framings. This board reports Overall chance-adjusted accuracy (%): raw single-select accuracy minus the per-item chance baseline, rescaled so random guessing scores 0 and a correct answer 1; a negative value is worse than chance; higher is better.
Metric: Overall chance-adjusted accuracy (%): raw single-select accuracy minus the per-item chance baseline, rescaled so random guessing scores 0 and a correct answer 1; a negative value is worse than chance; higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 58.52 | #232 |
| 2 | Claude Opus 4.6 | 55.13 | #102 |
| 3 | Qwen 3.5 397B A17B | 53.42 | #246 |
| 4 | GPT-5.4 | 50.28 | #123 |
| 5 | InternVL3-78B | 47.03 | #738 |
| 6 | Gemini 3.1 Pro (Preview) | 43.77 | #94 |
| 7 | Qwen2.5-Omni-7B | 30.71 | #1789 |
| 8 | InternVL3-8B | 29.49 | #1688 |
| 9 | Gemma 4 E4B | 28.67 | #2214 |
| 10 | Kimi K2.5 | 18.33 | #256 |
| 11 | Grok 4.1 | 13.71 | #476 |
| 12 | GLM-4.5V | 5.43 | #744 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=musebench-single-select-overall · How It Works · Data refreshed daily, snapshot 2026-11-13.