MUSE - Activity Localization: leaderboard
Locate the human activity shown in an artistic image, one of MUSE's three semantic-understanding tasks on artwork for situated language education. Accuracy, higher is better. Source: arxiv.org/abs/2609.19088.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 69.5 | #22 |
| 2 | InternVL3-38B | 61.5 | #413 |
| 3 | InternVL3-14B | 55 | #506 |
| 4 | GPT-4o | 51.5 | #357 |
| 5 | Qwen 2.5 VL 7B | 50 | #641 |
| 6 | Gemma 3 27B (IT) | 47 | #528 |
| 7 | InternVL3-8B | 44 | #624 |
| 8 | MiniCPM-V-2.6 | 43 | #827 |
| 9 | Gemma 3 12B (IT) | 39.5 | #672 |
| 10 | Gemma 3 4B (IT) | 32 | #997 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=muse-activity-localization · How It Works · Data refreshed daily, snapshot 2026-11-05.