whatmodel.sh / video / MAGI-1 4.5B
MAGI-1 4.5B
↗ share
+ contribute
MAGI-1 4.5B is an open-weights video model from Sand AI, released April 2025. Its published size is 4.5B parameters. Licence: apache-2.0.
Catalogue identity notes
Sand AI's MAGI-1 4.5B base checkpoint (ckpt/magi/4.5B_base; README row 'MAGI-1-4.5B'), released 30 Apr 2025; a 4.5B distill followed on 26 May 2025. PAI-Bench Table 12 lists 'MAGI-1-4.5B' at 720x720 (the README's default resolution for the 4.5B config), mapped to the base checkpoint named in the README table. Pinned sand-ai/MAGI-1@ca8bdfb. Comparison-only (README: runs on one RTX 4090; no single command verified).
Model information sources
Release date ↗ · Weights licence ↗
4.5B as named by Sand AI (model table 'MAGI-1-4.5B').
$ how it compares
The video models ranked either side of MAGI-1 4.5B, out of 139 with a score, then more from Sand AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 109 |  T2V-Turbo (VC2) | 38±3 | local | ||
| 110 |  Runway Gen-4 Turbo | 37±3 | api $0.050/sec | ||
| 111 | MAGI-1 4.5B | 37±5 | comparison only | ||
| 112 |  LTX-2.3 | 36±3 | api $0.040/sec +local | ||
| 113 |  CogVideoX1.5-5B | 36±2 | local | ||
| more from Sand AI | |||||
| 76 | MAGI-1 24B | 48±4 | comparison only | ||
$ what it's good at
image to video▁▁▃▂██▄▃▃▄▂▂ #68 of 74 * · 89% of its score · 2 tests
Beats Runway Gen-4 Turbo here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CRONOS counterfactual rigid-body physics (I2V success rate) | 2.0% ✎ | █·····█··█·█ beats 0/3 | 53% |
| PAI-Bench-G physical-AI video generation (Overall) | 77.5% ✎ | ▃······▅·█▅█ beats 1/10 | 36% |
$ where to run it
No verified API or download recorded yet · add one
Official weights and documentation: huggingface.co/sand-ai ↗
$ how the score was worked out · CRONOS: Benchmarking Counterf… 53%, PAI-Bench: A Comprehensive Be… 36%
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 |
|---|---|---|
 CRONOS: Benchmarking Counterfactual Physical Consistency in Video Models (arXiv:2605.23699v1), Table 1 | 1 | 53% |
| PAI-Bench: A Comprehensive Benchmark For Physical AI (arXiv:2512.01989v1) | 1 | 36% |
| prior · pull toward the middle | 11% | |
| weighted average | = 36.5 |
$ 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