whatmodel.sh / video / Cosmos-Predict2.5-14B
Cosmos-Predict2.5-14B
↗ share
+ contribute
Cosmos-Predict2.5-14B is an open-weights video model from NVIDIA, released December 2025. Its published size is 14B parameters. Licence: nvidia-open-model-license.
Catalogue identity notes
NVIDIA Cosmos-Predict2.5 14B base model, released 5 Dec 2025 (cosmos-predict2.5 README: 'Released Cosmos-Predict2.5-14B base models'); HF nvidia/Cosmos-Predict2.5-14B pinned 18839bf (NVIDIA Open Model License per card). PhyGround's 'Cosmos-14B' row is listed in its Appendix generation table as 'Cosmos-Predict2.5-14B' (93 frames, 16 fps, 1280x704); the Cosmos 3 report (NVIDIA) runs 'Cosmos-Predict2.5-14B' on Cosmos HUE. Comparison-only: no local command verified in this pass.
Model information sources
Release date ↗ · Weights licence ↗
14B as named by NVIDIA (model name). Text encoder and tokenizer not included.
$ how it compares
The video models ranked either side of Cosmos-Predict2.5-14B, out of 139 with a score, then more from NVIDIA.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 82 |  Kandinsky 5.0 T2V Lite | 47±2 | local | ||
| 83 |  LTX-2 19B Distilled | 47±8 | api $0.020/sec +local | ||
| 84 | Cosmos-Predict2.5-14B | 47±3 | comparison only | ||
| 85 |  CogVideoX-5B-I2V | 46±4 | local | ||
| 86 | Cosmos-1.0-Diffusion-7B-Text2World | 46±3 | local | ||
| more from NVIDIA | |||||
| 38 | Cosmos3-Super | 58±4 | local | ||
| 60 | Cosmos3-Nano | 52±3 | local | ||
| 72 | Cosmos-1.0-Diffusion-7B-Video2World | 49±5 | comparison only | ||
$ what it's good at
text to video▇▅▄▆█▆▇▂▃▃▂▄ #49 of 94 * · 32% of its score · 1 test
Beats Wan 2.2 TI2V 5B here, though they rank higher overall. Trails Cosmos-1.0-Diffusion-7B-Text2World here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Cosmos HUE human evaluation, text-to-video (Cosmos 3 report) lab | 82.1% ✎ | █▄▄··▄▄▄▄▄·▄ beats 2/9 | 32% |
image to video▁▁▂▂██▄▃▃▄▂▂ #59 of 74 * · 56% of its score · 2 tests
Beats Wan 2.2 A14B, LTX-2 here, though they rank higher overall. Trails Hailuo 01 (MiniMax video-01 / I2V-01) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CRONOS counterfactual rigid-body physics (I2V success rate) | 8.0% ✎ | █·····█··█·█ beats 1/3 | 31% |
| Cosmos HUE human evaluation, image-to-video (Cosmos 3 report) lab | 83.0% ✎ | ▄·█··▄··▄▄▄█ beats 2/8 | 25% |
other tests1 not in a task · 8.7% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| PhyGround physics human evaluation (Overall, TI2V) | 58.2% ✎ | █▄▄··▄···▄·█ beats 4/7 | 8.7% |
$ where to run it
No verified API or download recorded yet · add one
Official weights and documentation: huggingface.co/nvidia ↗
$ how the score was worked out · Cosmos 3: Omnimodal World Mod… 57%, CRONOS: Benchmarking Counterf… 31%, PhyGround: Benchmarking Physi… 9%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4) · lab-reported | 2 | 57% |
| CRONOS: Benchmarking Counterfactual Physical Consistency in Video Models (arXiv:2605.23699v1), Table 1 | 1 | 31% |
| PhyGround: Benchmarking Physical Reasoning in Generative World Models (arXiv:2605.10806v1) | 1 | 9% |
| prior · pull toward the middle | 4% | |
| weighted average | = 46.8 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json