Mochi 1
A solid pick for self-hosting and full data control.
Overview
Mochi 1 is Genmo's text-to-video diffusion model, released with open weights under Apache 2.0. Its 10-billion-parameter AsymmDiT architecture delivers strong prompt adherence and convincing photorealism, though only at 480p for now. The real limit: extreme motion can still warp footage, and the model struggles with animated or stylized styles. Because weights are fully open, it can run entirely on local hardware, keeping data in-house — appealing for organisations wary of cloud lock-in, provided you have the GPU memory to spare.
Skill profile
Not disclosed
Strengths
- Open-source and self-hostable
Limitations
- API pricing not disclosed
Who is it for
- you want to control cost or self-host
- you have GDPR constraints
Ideal use cases
- Marketing video prototyping
- AI-generated storyboards
- In-house video-gen R&D
- Rapid product demos
Access & availability
Key specifications
Privacy
Run it locally
| Quantization | Disk | RAM / VRAM | Typical hardware |
|---|---|---|---|
| Q4 · recommended | 5.7 GB | 8 GB | 16GB PC / M1+ Mac / 8GB GPU |
| Q8 · balanced | 10.7 GB | 14 GB | 32GB PC / 24GB Mac / RTX 4070 Ti+ |
| FP16 · max quality | 20 GB | 24 GB | 32GB Mac / RTX 3090-4090 |
Estimates for a moderate context. Long contexts need more RAM (KV cache).
Deploy
Copy-ready commands generated from this card. Adjust context length and GPU count to your hardware.
OpenAI-compatible server for production on NVIDIA GPUs.
pip install vllm
vllm serve genmo/mochi-1-preview \
--max-model-len 32768 \
--tensor-parallel-size 1 \
--dtype autoAdvanced data · for expertsArchitecture, modalities, detailed cost, full benchmarks▾
| Arena Elo | — |
| MMLU | — |
| GPQA | — |
| HumanEval | — |
| SWE-Bench | — |
| MATH | — |