MMOOC - Multimodal Ambiguity (Open-Ended VQA): leaderboard
Metric: Out-of-context score (%; mean of the refusal rate, the share of the Multimodal Ambiguity Open-Ended VQA questions, which the image and context cannot answer, that the model correctly declines, and refusal rationality, the mean 0-1 judge score of its abstention reasoning in steps of 0.25; judge scores averaged over GPT-5.6, Claude Opus 5 and DeepSeek-V4-Pro). Source: arxiv.org. Saturation forecast: Around 2030. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemma 4 26B | 79.25 |
| 2 | O1 | 75 |
| 3 | Gemma 4 31B | 71.5 |
| 4 | Ministral 3 8B | 54.5 |
| 5 | Ministral 3 14B | 52.5 |
| 6 | Llama 4 Maverick | 52.25 |
| 7 | O3 | 47.75 |
| 8 | Qwen 3.5 27B | 46 |
| 9 | Qwen 3.5 122B A10B | 46 |
| 10 | GPT-4o | 43.5 |
| 11 | InternVL3-8B | 21.75 |
| 12 | Claude Opus 4.6 | 14.75 |
| 13 | Gemini 3.1 Pro (Preview) | 7.75 |
Interactive version: theaggregate.ai/benchmark?slug=mmooc-multimodal-ambiguity-open-ended-vqa · How It Works · Data refreshed daily, snapshot 2026-09-29.