All tasks / Models
MAGI-2 Preview
Sand.ai's open-weights 114B MoE (6B active) audio-video model generating 10 s clips; needs a multi-GPU server.
- Sand.ai
- Open weights
- Released 5 Aug 2026
Where it ranks
- BVideo generationModels · Local · 1,045 on Artificial Analysis Video Arena: Text to Video (AA-Video-T2V v2.1, with audio)
Facts
- Vendor
- Sand.ai
- Family
- MAGI
- Released
- 5 Aug 2026
- Weights
- Open: download and run it yourself
- License
- Apache-2.0
- Parameters
- 114B (6B active)
- GPU memory to run locally
- ~640 GB (README requirement: 8 NVIDIA Hopper GPUs (80 GB each); checkpoints total ~307 GB)
- Access
- Free (open weights)
- Last verified
- 18 Oct 2026
Price from huggingface.co, 18 Oct 2026.
Benchmark results
| Benchmark | Result | Entry on the leaderboard | As of | Source |
|---|---|---|---|---|
| Artificial Analysis Video Arena: Image to Video (AA-Video-I2V v1.1, with audio) | 1,046 | MAGI-2 PreviewAA-Video-I2V v1.1 (with audio) board | 18 Oct 2026 | artificialanalysis.ai |
| Artificial Analysis Video Arena: Text to Video (AA-Video-T2V v2.1, with audio) | 1,045 | MAGI-2 PreviewAA-Video-T2V v2.1 (with audio) board | 18 Oct 2026 | artificialanalysis.ai |
| Artificial Analysis Video Arena: Text to Video, no audio (AA-Video-T2V-Silent v2.1) | 1,012 | MAGI-2 PreviewAA-Video-T2V-Silent v2.1 (no audio) board | 18 Oct 2026 | artificialanalysis.ai |
Where you can use it
No tracked tool lists this model yet; as open weights it runs in local runtimes.
Earlier versions
- MAGI-1 · Apr 2025