CoMMET (Image-Dependent): leaderboard
Metric: Overall story accuracy (%, a StoryTurn counts only if every question in the multi-turn conversation is answered correctly), open-ended answers judged against the ground truth by Gemini 3.1 Pro, on the 192 image-dependent StoryTurns of CoMMET (answering needs the illustrations), stories with images; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 62.5 | #45 |
| 2 | Gemini 3.1 Flash Lite | 57.29 | #174 |
| 3 | GPT-5.4 | 53.65 | #76 |
| 4 | GPT-4o | 48.44 | #333 |
| 5 | Claude Haiku 4.5 | 45.83 | #271 |
| 6 | Llama 4 Maverick | 35.42 | #451 |
| 7 | Llama 3.2 11B Instruct | 17.19 | #1112 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=commet-image-dependent · How It Works · Data refreshed daily, snapshot 2026-10-11.