whatmodel.sh / video / MAGI-1 24B
MAGI-1 24B
↗ share
+ contribute
MAGI-1 24B is an open-weights video model from Sand AI, released April 2025. Its published size is 24B parameters. Licence: apache-2.0.
Catalogue identity notes
Sand AI's MAGI-1 autoregressive video model, 24B base checkpoint (repo folder ckpt/magi/24B_base; README model table row 'MAGI-1-24B'), released 21 Apr 2025. Pinned sand-ai/MAGI-1@ca8bdfb. PAI-Bench (Dec 2025) lists 'MAGI-1-24B' at 120 frames, 24 fps, 1280x720. The repo also holds a 24B distill, an fp8 distill and, uploaded in June 2026, MAGI-1.1 24B weights; the Dec-2025 PAI-Bench row predates 1.1, so it is mapped here. The WROP (Sep 2026) 'MAGI-1 24B' row is NOT mapped because it could be the 1.1 upload (held). Comparison-only: no single-command local route verified (official route needs 8xH100 or 8xRTX 4090 per README).
Model information sources
Release date ↗ · Weights licence ↗
24B as named by Sand AI (model table 'MAGI-1-24B'); T5 text encoder and VAE separate.
$ how it compares
The video models ranked either side of MAGI-1 24B, out of 139 with a score, then more from Sand AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 74 |  Cosmos-Predict2.5-2B | 48±2 | comparison only | ||
| 75 | Cosmos-Predict2-14B-Video2World | 48±8 | comparison only | ||
| 76 | MAGI-1 24B | 48±4 | comparison only | ||
| 77 |  Kandinsky 6.0 Video Pro | 48±7 | local | ||
| 78 |  Wan2.1-I2V-14B-720P | 48±4 | comparison only | ||
| more from Sand AI | |||||
| 111 | MAGI-1 4.5B | 37±5 | comparison only | ||
$ what it's good at
image to video▁▁▂▂██▄▃▃▄▂▂ #54 of 74 * · 92% of its score · 2 tests
Beats HunyuanVideo 1.5, Wan 2.2 A14B, LTX-2 here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| PAI-Bench-G human preference arena (Overall ELO) | 988.9 ✎ | ▄······██··█ beats 3/6 | 69% |
| PAI-Bench-G physical-AI video generation (Overall) | 79.4% ✎ | ▃······▅·█▅█ beats 3/10 | 23% |
$ 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 · PAI-Bench: A Comprehensive Be… 92%
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 |
|---|---|---|
 PAI-Bench: A Comprehensive Benchmark For Physical AI (arXiv:2512.01989v1) | 2 | 92% |
| prior · pull toward the middle | 8% | |
| weighted average | = 48.3 |
$ 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