CulMind - Seals: leaderboard
CulMind (2026) tests multimodal models on 50 Chinese cultural heritage tasks built from more than 100 museum collections. This board is the seals subdomain score, the mean of its task scores on a 0 to 100 scale in the answer-only setting; higher is better.
Metric: Seals subdomain score: macro-average of the task scores (0-100) in the subdomain, each of the 50 tasks keeping its own primary metric such as accuracy or F1, answer-only setting, images from more than 100 museum collections; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 32B Instruct | 53.1 | #585 |
| 2 | Gemini 3 Flash (Preview) | 51.5 | #144 |
| 3 | Qwen 3 VL 8B Instruct | 48.3 | #981 |
| 4 | GPT-5.5 | 46.5 | #49 |
| 5 | Gemini 2.5 Flash | 45.4 | #453 |
| 6 | GLM-4.5V | 45.2 | #744 |
| 7 | InternVL3-8B | 42.6 | #1688 |
| 8 | InternVL3-14B | 41.6 | #1182 |
| 9 | Qwen 3.5 27B | 41 | #359 |
| 10 | GPT-5.4 Mini | 39.7 | #389 |
| 11 | Qwen 3.5 9B | 37.1 | #803 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=culmind-seals · How It Works · Data refreshed daily, snapshot 2026-11-13.