Dia vs Kyutai Moshi

A side-by-side look at pricing, capabilities, pros, cons, and our editorial scores.

	Dia Audio	Kyutai Moshi Audio
Tagline	Open-weights 1.6B text-to-dialogue model that generates ultra-realistic multi-speaker conversations in one pass.	Open-source, full-duplex speech-to-speech foundation model with sub-200ms latency
Category	Audio	Audio
Pricing	Free· Free, open weights (Apache 2.0); hosted larger version waitlisted	Free· Free and open source. Models under CC-BY 4.0, code under MIT (Python) / Apache 2.0 (Rust). Self-hosted only — you pay your own compute (24GB+ GPU for PyTorch, or Apple Silicon via MLX).
Model	Dia-1.6B	Moshi (7B-class speech-text foundation model) + Mimi neural audio codec, in-house by Kyutai
Editorial score	7.3 / 10	—
Use cases	dialogue-generationvoice-cloningpodcast-prototypinggame-voice-actingtext-to-speech	Real-time voice assistant prototypesResearch on full-duplex spoken dialogueOn-device voice interaction on Apple Silicon via MLXLow-latency conversational agents behind WebSocketNeural audio codec experimentation with MimiSelf-hosted voice interface for privacy-sensitive appsSpeech tokenizer for downstream audio LLM trainingInterruptible in-car or wearable voice UX
Pros	Open weights under Apache 2.0 with first-party Transformers support Multi-speaker [S1]/[S2] dialogue and nonverbal tags in a single pass Zero-shot voice cloning from a short audio prompt plus transcript Runs ~2x realtime on a single RTX 4090 at ~4.4GB VRAM Free Hugging Face ZeroGPU Space to try without local GPU	Truly full-duplex — handles interruptions, overlap and back-channels rather than rigid turn-taking Sub-200ms practical latency on a single L4 GPU, well below third-party voice APIs Fully open weights (CC-BY 4.0) plus MIT/Apache code — self-host with no per-minute billing Ships with Mimi, a streaming neural audio codec that beats SpeechTokenizer and SemantiCodec Multiple inference backends: PyTorch for research, Rust/Candle for production, MLX for on-device Mac/iPhone Inner-monologue text prediction gives you a transcript alongside the audio stream for free
Cons	English only; no built-in multilingual support Voices drift between runs unless you fix a seed or supply a prompt GPU required; CPU inference not yet supported Tiny team (1.5 engineers); slower issue turnaround than commercial TTS	English-only voices at launch — no multilingual support out of the box Knowledge and reasoning quality trail top text LLMs; it's a 7B-class model, not GPT-4o Voice Requires a 24GB+ GPU for the reference PyTorch build; on-device is only viable via MLX on Apple Silicon No hosted API or SaaS tier — you own the ops, scaling and safety filtering Only two fixed synthetic voices (Moshiko/Moshika); no voice cloning or speaker conditioning in the release
Website	github.com	kyutai.org

Pick Dia if

✅ Open weights under Apache 2.0 with first-party Transformers support
✅ Multi-speaker [S1]/[S2] dialogue and nonverbal tags in a single pass
✅ Zero-shot voice cloning from a short audio prompt plus transcript
✅ Runs ~2x realtime on a single RTX 4090 at ~4.4GB VRAM

Pick Kyutai Moshi if

✅ Truly full-duplex — handles interruptions, overlap and back-channels rather than rigid turn-taking
✅ Sub-200ms practical latency on a single L4 GPU, well below third-party voice APIs
✅ Fully open weights (CC-BY 4.0) plus MIT/Apache code — self-host with no per-minute billing
✅ Ships with Mimi, a streaming neural audio codec that beats SpeechTokenizer and SemantiCodec

Compare a different pair →