Physics-IQ Verified
Dynamic ranking of video models by how well they understand physical principles.
41models16labs
Sampling
Includes benchmarked models and models known from preprints or announcements, even if public access is unavailable or no release date is confirmed.
| RANK | MODEL | NET IMPROVEMENTVS TRACK MEAN · PP | LLM USAGE | PROMPT SOURCE | OUTPUT FPS | COST / VIDEO Normalized cost |
|---|---|---|---|---|---|---|
| 01 |  FLUX 3 [large]Black Forest Labs · PROPRIETARY | +12.27 pp± 0.44 pp | Yes | Custom | 24 | $0.868 |
| 02 |  Odyssey-3 ProOdyssey · PROPRIETARY | +11.15 pp± 0.42 pp | Yes | Custom | 16 | $0.267 |
| 03 | Odyssey-3Odyssey · PROPRIETARY | +9.99 pp± 0.78 pp | Yes | Custom | 16 | $0.139 |
| 04 |  Physis-Lang (Cosmos3 Super)NVIDIA · OPEN SOURCE | +9.39 pp± 1.43 pp | Yes | Custom | 24 | $0.823 |
| 05 | Physis-Lang (Cosmos3 Nano)NVIDIA · OPEN SOURCE | +4.45 pp± 1.52 pp | Yes | Custom | 24 | $0.823 |
| 06 | Cosmos3 SuperNVIDIA · OPEN SOURCE | +3.84 pp± 0.76 pp | Yes | Custom | 24 | $0.823 |
| 07 |  Seedance 2.5Seedance · PROPRIETARY | +3.59 pp± 0.76 pp | Yes | BPP | 24 | $2.838 |
| 08 | Odyssey-3Odyssey · PROPRIETARY | +2.15 pp± 0.84 pp | No | BPP | 16 | $0.129 |
| 09 |  MiniMax H3MiniMax · OPEN SOURCE | +0.95 pp± 0.35 pp | Yes | BPP | 24 | $0.438 |
| 10 | Cosmos3 NanoNVIDIA · OPEN SOURCE | -1.59 pp± 0.89 pp | Yes | Custom | 24 | $0.434 |
| 11 |  MiniMax H3 Maxfal · PROPRIETARY | -2.63 pp± 0.67 pp | Yes | BPP | 24 | $0.482 |
| 12 |  Gemini Omni 1.1 FlashGoogle · PROPRIETARY | -3.50 pp± 0.44 pp | Yes | BPP | 24 | $0.609 |
| 13 |  Grok Imagine VideoxAI · PROPRIETARY | -4.04 pp± 0.60 pp | Yes | BPP | 24 | $0.352 |
| 14 |  Hunyuan Video 1.5Tencent · OPEN SOURCE | -5.44 pp± 0.80 pp | Yes | BPP | 24 | $0.604 |
| 15 | Gemini Omni FlashGoogle · PROPRIETARY | -5.48 pp± 0.19 pp | Yes | BPP | 24 | $0.609 |
| 16 | Cosmos3 EdgeNVIDIA · OPEN SOURCE | -6.10 pp± 1.05 pp | Yes | BPP | 24 | n.d. |
| 17 |  Wan 2.2 14BAlibaba · OPEN SOURCE | -6.64 pp± 0.60 pp | No | BPP | 16 | $0.165 |
| 18 | Veo 3.1 LiteGoogle · PROPRIETARY | -7.01 pp± 0.32 pp | Yes | BPP | 24 | $0.300 |
| 19 |  CogVideoX-5BZ.ai · OPEN SOURCE | -7.09 pp± 1.47 pp | No | BPP | 8 | $1.256 |
| 20 |  Kandinsky-WM 1.0Kandinsky Lab · OPEN SOURCE | -8.02 pp± 0.86 pp | Yes | BPP | 24 | $0.222 |
| 21 | Veo 3.1 FastGoogle · PROPRIETARY | -8.88 pp± 0.53 pp | Yes | BPP | 24 | $0.600 |
| 22 | Wan 2.2 5BAlibaba · OPEN SOURCE | -11.13 pp± 0.91 pp | No | BPP | 24 | $0.090 |
| 23 |  Sora 2OpenAI · PROPRIETARY | -12.34 pp± 0.80 pp | Yes | BPP | 30 | $0.640 |
| 24 |  P-VideoPruna AI · PROPRIETARY | -13.54 pp± 1.80 pp | No | BPP | 24 | $0.100 |
COST FRONTIER
Score vs Cost ($)
Cost uses a logarithmic x-axis to keep low-cost models legible. Generation cost is normalized to 24 FPS and 1280-wide output where those inputs are reported; otherwise the reported cost is retained. Separate LLM/prompt costs are added after generation-cost normalization. Odyssey API prompt rewriting is estimated at $0.01 per submitted video; each BoN×8 run shares one rewritten prompt, and selection compute is excluded.
METRIC BREAKDOWN
Ranked independently · verified scoresSubmetric leaders
SPSpatial
| RANK | MODEL | SCORE |
|---|---|---|
| 01 | FLUX 3 [large] | 64.36 ± 0.59 |
| 02 | Odyssey-3 Pro | 61.78 ± 0.40 |
| 03 | Physis-Lang (Cosmos3 Super) | 59.87 ± 1.43 |
| 04 | Odyssey-3 | 59.17 ± 1.01 |
| 05 | MiniMax H3 | 58.90 ± 0.78 |
STSpatiotemporal
| RANK | MODEL | SCORE |
|---|---|---|
| 01 | Odyssey-3 | 44.70 ± 1.86 |
| 02 | Odyssey-3 Pro | 42.97 ± 1.78 |
| 03 | Physis-Lang (Cosmos3 Super) | 41.57 ± 3.11 |
| 04 | FLUX 3 [large] | 41.27 ± 0.99 |
| 05 | Odyssey-3 | 37.40 ± 1.25 |
WSWeighted spatial
| RANK | MODEL | SCORE |
|---|---|---|
| 01 | FLUX 3 [large] | 53.25 ± 0.91 |
| 02 | Odyssey-3 Pro | 51.26 ± 0.36 |
| 03 | Physis-Lang (Cosmos3 Super) | 48.53 ± 1.06 |
| 04 | Odyssey-3 | 48.20 ± 1.30 |
| 05 | Physis-Lang (Cosmos3 Nano) | 43.39 ± 2.16 |
MSEMean squared error
| RANK | MODEL | SCORE |
|---|---|---|
| 01 | FLUX 3 [large] | 45.55 ± 0.55 |
| 02 | Odyssey-3 Pro | 43.96 ± 0.91 |
| 03 | Odyssey-3 | 43.24 ± 0.57 |
| 04 | Physis-Lang (Cosmos3 Super) | 42.96 ± 1.29 |
| 05 | Physis-Lang (Cosmos3 Nano) | 42.58 ± 0.85 |