# Open weights — page 3

> Open-weights model releases and what people are doing with them.

- [k2-fsa releases OmniVoice, a 0.6B zero-shot TTS covering 600+ languages](https://shipped.anth.us/open-weights/k2-fsa-omnivoice/index.md): OmniVoice is a 0.6B voice-cloning and voice-design TTS tagged for 646 languages; code Apache 2.0, weights CC-BY-NC. (2026-03-30)
- [Mistral releases Voxtral TTS, a 4B open-weights speech model, non-commercial](https://shipped.anth.us/open-weights/mistral-voxtral-tts/index.md): Voxtral 4B TTS covers nine languages with 20 preset voices and streaming; weights are CC BY-NC 4.0. (2026-03-26)
- [Google DeepMind releases Gemma 4, with a 31B dense model under Apache 2.0](https://shipped.anth.us/open-weights/google-gemma-4-31b-it/index.md): Gemma 4 comes in five sizes from E2B to 31B, with multimodal input, 256K context on the larger models and an Apache 2.0 licence. (2026-03-11)
- [Fish Audio releases S2 Pro, a 5B TTS model for 80+ languages, research licence](https://shipped.anth.us/open-weights/fish-audio-s2-pro/index.md): S2 Pro is a dual-autoregressive TTS model with inline emotion tags; weights are free for research, commercial use needs a licence. (2026-03-09)
- [NineNineSix releases KaniTTS2-en, a 400M English TTS with voice cloning](https://shipped.anth.us/open-weights/nineninesix-kani-tts-2-en/index.md): KaniTTS2-en is a 400M English TTS on LFM2 350M with cloning via a speaker embedder, about 3 GB VRAM; LFM 1.0 licence. (2026-02-12)
- [OpenMOSS Team releases MOSS-TTS, an 8B Apache 2.0 voice cloning and long-form TTS](https://shipped.anth.us/open-weights/openmoss-moss-tts/index.md): MOSS-TTS is an 8B (and 1.7B) open TTS family with cloning, up to an hour of long-form audio and duration control. (2026-02-06)
- [Qwen releases Qwen3-TTS, Apache 2.0 voice cloning in 10 languages](https://shipped.anth.us/open-weights/alibaba-qwen3-tts/index.md): Qwen3-TTS is a family of open TTS models, 0.6B and 1.7B, with 3-second voice cloning, voice design and streaming, under Apache 2.0. (2026-01-22)
- [Hume AI releases TADA-1B, an English TTS that aligns one speech vector per text token](https://shipped.anth.us/open-weights/humeai-tada-1b/index.md): TADA-1B is Hume AI's open English TTS built on Llama 3.2 1B, with 1:1 text-to-speech alignment; Llama 3.2 licence. (2026-01-12)
- [Soprano-80M: an 80M Apache 2.0 TTS that claims 2000x real time on GPU](https://shipped.anth.us/open-weights/ekwek-soprano-80m/index.md): Soprano-80M is a tiny English-only TTS with under 15 ms streaming latency on GPU, no cloning, Apache 2.0; superseded by Soprano 1.1. (2025-12-22)
- [FunAudioLLM releases Fun-CosyVoice3-0.5B-2512, an Apache 2.0 voice cloning TTS](https://shipped.anth.us/open-weights/funaudio-fun-cosyvoice3-0-5b-2512/index.md): Fun-CosyVoice3 is a 0.5B zero-shot cloning TTS for nine languages and 18+ Chinese dialects, with latency down to 150 ms. (2025-12-11)

Page 3 of 4 · [previous](https://shipped.anth.us/open-weights/page/2/index.md) · [next](https://shipped.anth.us/open-weights/page/4/index.md)