{"id":"pocket-tts-4-0","slug":"pocket-tts-4-0","name":"Pocket TTS 4.0","orgId":"kyutai","family":"pocket-tts-4-0","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://x.com/kyutai_labs/status/2112554656459084053","hfRepo":null,"modalities":["audio"],"aliases":["Pocket TTS 4.0","pocket tts 4.0","pocket-tts 4.0","pocket tts v4.0.0","pocket-tts v4.0.0","english_18m_2026-10","english_18m_refl_2026-10","english_7m_2026-10","english_7m_refl_2026-10","english_2m_2026-10","english_2m_refl_2026-10"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-20","announced":"2026-10-20","blurb":"Kyutai's Pocket TTS v4.0.0 release: six small English text-to-speech models with voice cloning, at 18.2M, 6.9M and 1.9M parameters, each with a reward-fine-tuned variant. Weights, training and inference code are open; the smallest model is reported to run about 103× faster than real time on one CPU core. Not an LLM.","api":"none","openWeights":"yes","params":{"text":"18.2M, 6.9M and 1.9M (decoder included)","source":"Kyutai's pocket-tts README on GitHub"},"facts":[{"label":"Models","value":"english_18m_2026-10, english_7m_2026-10 and english_2m_2026-10 (experimental), each with a _refl variant that adds reward fine-tuning"},{"label":"Earlier Pocket TTS","value":"About 100M parameters; still available. The new 18.2M model has the original's quality (lab statement)"},{"label":"Voice cloning","value":"Yes, in all six models"},{"label":"Code","value":"MIT (kyutai-labs/pocket-tts); training code public since August 2026; v4.0.0 also ships prebuilt runtime binaries"},{"label":"Weights","value":"Hugging Face kyutai/pocket-tts, CC BY 4.0, access requires accepting the terms"}],"claims":["Lab claim: the 1.9M model runs 103× faster than real time on one CPU core, and the 18.2M model matches the quality of the original Pocket TTS."],"sources":[{"kind":"lab","title":"Kyutai post on X with the new model sizes","url":"https://x.com/kyutai_labs/status/2112554656459084053","date":"2026-10-20"},{"kind":"lab","title":"Kyutai announcement on X","url":"https://x.com/kyutai_labs/status/2112565301816475970","date":"2026-10-20"},{"kind":"lab","title":"Post on X by Kyutai's Alexandre Défossez","url":"https://x.com/honualx/status/2112568675878486460","date":"2026-10-20"},{"kind":"repo","title":"pocket-tts v4.0.0 release on GitHub","url":"https://github.com/kyutai-labs/pocket-tts/releases/tag/v4.0.0","date":"2026-10-20"},{"kind":"repo","title":"kyutai/pocket-tts on Hugging Face","url":"https://huggingface.co/kyutai/pocket-tts","date":"2026-10-20"}],"firstBenchmark":null},"organization":{"id":"kyutai","name":"Kyutai","slug":"kyutai","country":null,"url":null,"color":"#11110F","mark":"K"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 20, 2026; no independent results yet.","lead":"As of Oct 22, 2026, Pocket TTS 4.0 is pending benchmarks on isitsota: announced Oct 20, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}