WMW (World Models in Words): leaderboard
Vision-language models (VLMs) are increasingly used to answer questions about physical scenes, yet most evaluations reduce performance to a final answer. This hides whether the model perceived the right objects, represented the right physical state, predicted a plausible transition, or merely selected the right option for the wrong reasons. We introduce WMW, an evaluation framework for auditing
Metric: Trace–ans. consistency (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 91 | #49 |
| 2 | GPT-5.5 | 88 | #31 |
| 3 | GPT-4o | 83 | #343 |
| 4 | InternVL3-78B | 80 | #352 |
| 5 | GPT-4o Mini | 78 | #599 |
| 6 | Qwen 2.5 VL 7B | 72 | #643 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=wmw-world-models-in-words · How It Works · Data refreshed daily, snapshot 2026-10-20.