Mochi 1
Video Deprecated Full comparison ↗ genmo/mochi-1 · by Genmo AI
· Diffusion transformer (DiT)
Pricing — 1 offering(s)
About this offering: Genmo's open-weights Mochi 1 text-to-video model, hosted on Replicate (genmoai/mochi-1). It's billed by compute time on an Nvidia H100 at $0.001525 per second; a clip typically takes up to 5 minutes, and a median run costs about $0.42 (Replicate model page, checked 2026-10-11). Genmo no longer develops Mochi 1.
Standard
- $0.00153 / GPU second
Current
2026-10-18 → present
Billed per second of GPU time on an Nvidia H100, not per second of output video. A typical 5-second clip…
Billed per second of GPU time on an Nvidia H100, not per second of output video. A typical 5-second clip costs about $0.42 (Replicate's median run cost; an estimate that varies with inputs). Runs typically take several minutes.
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Mochi 1 fits into a cost-aware routing setup
See how →Capability profile
How Mochi 1 rates across core capability dimensions, with the task-level evidence behind each rating.
Was among the strongest open-weight models for motion quality at launch (Oct 2024); now superseded by Wan 2.x and HunyuanVideo.
No audio.
Max 480p; not suited for production use requiring HD output.
~5 minutes per clip on Replicate; very slow for production use.
Max ~6 seconds. Short by current standards.
T2V only (no I2V); no editing of existing clips.
Operator guidance
Mochi 1 is deprecated; do not use for new production deployments. Included for historical benchmarking and comparison. Self-hosting is possible (Apache-2.0 weights), but current open-weight alternatives (Wan 2.5, HunyuanVideo 1.5, LTX Video) significantly outperform it on all axes. On Replicate at $0.42/clip it is expensive relative to its quality tier.
Use cases
- Historical benchmarking and research comparisons
- Self-hosted experimentation with open-weight video DiTs
Limitations
- Deprecated; not under active development by Genmo
- Max 480p resolution; not suitable for HD output
- Slow inference (~5 min/clip on Replicate A100)
- Superseded by multiple open-weight alternatives
- $0.42/clip on Replicate is poor value given quality tier