HELM VHELM - Mme - Artwork: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) MME artwork subtask: yes/no questions about recognizing artworks in images; 400 questions, prefix quasi-exact match. Percent correct, 0-100, higher is better; guessing scores 50.
Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 (2025-04-16) | 87.5 | #125 |
| 2 | Claude 3.5 Sonnet (20241022) | 87.25 | #302 |
| 3 | Claude 3.5 Sonnet (20240620) | 86.25 | #308 |
| 4 | GPT-4.1 (2025-04-14) | 86 | #257 |
| 5 | Gemini 1.5 Pro (002) | 85.75 | #368 |
| 6 | GPT-4.5 | 85.5 | #278 |
| 7 | Gemini 1.5 Flash (002) | 85.25 | #587 |
| 8 | Gemini 2.5 Pro (Preview 03-25) | 84.5 | #126 |
| 9 | Gemini 2.0 Pro (Preview 02-05) | 83.75 | #318 |
| 10 | Llama 4 Scout Instruct | 82.75 | #547 |
| 11 | GPT-4o (2024-11-20) | 82.5 | #392 |
| 12 | Gemini 2.0 Flash | 82 | #351 |
| 13 | GPT-4o (2024-08-06) | 80.75 | #343 |
| 14 | O1 (2024-12-17) | 80.75 | #157 |
| 15 | GPT-4 Turbo | 80.5 | #417 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-mme-artwork · How It Works · Data refreshed daily, snapshot 2026-11-05.