Kokoro vs Pocket TTS: which tiny open TTS model should you use?
Kokoro has 54 fixed voices in 8 languages under Apache 2.0; Pocket TTS clones a voice from 5 seconds of audio in 6 languages under CC BY 4.0. Both run on a CPU.
Verdict. Choose by whether you need voice cloning. Kokoro (82M, 327 MB) has 54 fixed voices in eight languages, no cloning and a plain Apache 2.0 license, and it is by far the most downloaded speech model on this site, but its author is an independent developer who has not released a new model since February 2025. Pocket TTS (100M) clones a voice from about five seconds of audio in six languages and has kept improving, with much smaller English models published in October 2026. Its weights are listed as CC BY 4.0 behind a prohibited-use gate, while the launch post says MIT, and each preset voice has its own license. In Kyutai's tests Pocket TTS had the lowest word error rate (1.84%), but its speaker similarity trailed the other voice-cloning models and its audio quality trailed Chatterbox Turbo. Kyutai reports that both run faster than real time on a laptop CPU.
Specifications
| Specifications | Kokoro | Pocket TTS |
|---|---|---|
| Developer | hexgrad | Kyutai |
| Released | January 27, 2025 | January 13, 2026 |
| Total parameters | 82M (Kokoro-82M v1.0) | 100M (Pocket TTS 100M (six languages)) |
| Active parameters | All | All |
| Architecture | Other | Other |
| Context window | – | – |
| Input | Text | Text, Audio |
| License | Apache 2.0 (Permissive) | CC BY 4.0 (Permissive) |
| Memory for the smallest variant (4-bit) | 0.38 GB (Kokoro-82M v1.0) | 0.127 GB (Pocket TTS 18M (English only)) |
Performance claims come from the developer and have not been independently verified.
Pick Kokoro if
- You want a plain Apache 2.0 license with no access gate, and fixed voices are enough.
- You need Japanese, Mandarin Chinese or Hindi, which Pocket TTS does not list and Kokoro covers with ready-made voices.
- You want the model with the widest community use: about 12.5 million downloads in the 30 days to 11 November 2026.
- You want a small 327 MB checkpoint.
Pick Pocket TTS if
- You want to clone a voice from about five seconds of audio, with the speaker's consent.
- You need German or Dutch, which Kokoro does not cover.
- You want a project that is still being developed: Kyutai added per-language models, training code and tiny English models during 2026.
- You want an even smaller English model: Kyutai's 18.2M version runs at 9.9 times real time on one CPU thread, by its own measurement.