Arcadia MMMU Open Ended: leaderboard
Arcadia Impact inspect_ai evaluation of MMMU open-ended validation questions, testing multimodal academic reasoning with free-form answers.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash (001) | 39.62 | #360 |
| 2 | GPT-4o (2024-08-06) | 33.96 | #336 |
| 3 | Qwen 2.5 VL 72B Instruct | 33.96 | #377 |
| 4 | Claude 3.7 Sonnet (20250219) | 33.96 | #201 |
| 5 | Llama 3.2 90B Vision Instruct | 32.08 | #622 |
| 6 | DeepSeek R1 | 26.42 | #251 |
| 7 | GPT-4o Mini (2024-07-18) | 26.42 | #558 |
| 8 | Llama 3.3 70B Instruct | 20.75 | #534 |
| 9 | O1 Mini (2024-09-12) | 16.98 | #403 |
| 10 | DeepSeek V3 Chat | 16.98 | #447 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=arcadia-mmmu-open-ended · How It Works · Data refreshed daily, snapshot 2026-10-20.