K2V Study - PhyGenBench Thermodynamics: leaderboard
The K2V paper evaluates video-generation models and explicit generation pipelines on physical plausibility. PhysicsIQ and PhyGen metrics retain their own scales and cohorts. Training-free keyframe pipelines and centre-frame ablations remain separate configurations of their underlying models. This board reports phyGenEval PCA (0-1). Higher is better.
Metric: PhyGenEval PCA (0-1). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked. Updated 14 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | K2V+Wan2.2-14B | Evaluation run: K2V Gemini 3.8 Flash plus Gemini 3.1 Flash Image | 0.589 | |
| 2 | K2V+Wan2.2-5B | Evaluation run: K2V Gemini 3.8 Flash plus Gemini 3.1 Flash Image | 0.578 | |
| 3 | Wan2.2-TI2V-5B | 0.533 | ||
| 4 | Open-Sora V1.2 (1.1B) | 0.44 | ||
| 5 | Vchitect 2.0 (2B) | 0.44 | ||
| 6 | Wan2.2-T2V-A14B | 0.433 | ||
| 7 | Cosmos-Predict2.5 | 0.411 | ||
| 8 | CogVideoX-5B | 0.4 |
Interactive version: theaggregate.ai/benchmark?slug=k2v-study-phygenbench-thermodynamics · How It Works · Data refreshed daily, snapshot 2026-11-16.