CausalPhys - Entity Faithfulness: leaderboard
CausalPhys (2026) tests vision-language models on physical reasoning questions over real-world images and videos, each paired with an expert causal graph of objects, attributes and events. This board is the share in percent of causal-graph entities that the model rationale mentions, checked by a GPT-4o judge and averaged over the four domains; higher is better.
Metric: Entity Faithfulness (0-1 scaled to %), the share of causal-graph entities the rationale mentions, averaged over the four domains; the model writes a rationale and an answer; a GPT-4o judge makes binary checks of the rationale against the expert-annotated causal graph (typed object, attribute and event nodes with directed dependencies); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 69.68 | #738 |
| 2 | GPT-4o | 67.95 | #741 |
| 3 | GPT-4o Mini | 66.52 | #1623 |
| 4 | Claude Sonnet 4 | 66.34 | #400 |
| 5 | Gemini 2.5 Flash | 65.97 | #453 |
| 6 | Mistral Small 3.2 | 62.87 | #1591 |
| 7 | Phi-4 Multimodal Instruct | 62.26 | #3882 |
| 8 | Qwen 2 VL 7B | 58.71 | #2261 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=causalphys-entity-faithfulness · How It Works · Data refreshed daily, snapshot 2026-11-13.