All tasks / Models
Moshi (moshiko)
Kyutai's open full-duplex speech-to-speech dialogue model that listens and speaks at the same time.
- Kyutai
- Open weights
- Released Sep 2024
Where it ranks
- –Voice agentsModels · Local · no result on this task’s benchmarks yet
Facts
- Vendor
- Kyutai
- Family
- Moshi
- Released
- Sep 2024
- Weights
- Open: download and run it yourself
- License
- CC-BY-4.0 (weights); code Apache-2.0/MIT
- Parameters
- 7.69B
- GPU memory to run locally
- ~24 GB (Derived: 7.69B parameters (Hugging Face safetensors count, bf16 checkpoint) x 2 bytes = ~15.4 GB of weights plus runtime memory; int8 and MLX builds need less.)
- Access
- Free (open weights)
- Last verified
- 18 Oct 2026
Benchmark results
| Benchmark | Result | Entry on the leaderboard | As of | Source |
|---|---|---|---|---|
| Artificial Analysis Big Bench Audio (speech reasoning) | 4.4% | MoshiBig Bench Audio speech reasoning (audio in, audio out); embedded value 0.044 as percent; no last-updated date shown. | 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.