Kyutai

Kyutai is a Paris open-science AI lab whose Moshi speech-to-speech model brought real-time, full-duplex voice conversation to the open-source world.

Overview

Founded with backing from French tech donors, Kyutai develops multimodal models and releases them openly along with papers and benchmarks. Its flagship Moshi model listens and speaks simultaneously with natural turn-taking and emotional expression, and the lab has extended the line with text-to-speech (MoshiSpeech/TTS) and voice-activity tools. For builders, Kyutai offers hosted demos and inference code, making cutting-edge conversational voice AI accessible outside Big Tech labs.

Key Features

• Moshi: real-time full-duplex speech-to-speech conversation model
• Open weights and code for research and commercial experimentation
• Text-to-speech with natural prosody and expressive delivery
• Low-latency inference designed for live voice interaction
• Published papers, benchmarks and evaluation methodology
• Hosted playgrounds to try models without setup

Pricing

Kyutai releases its models openly at no cost; hosted API offerings for production use are expanding with usage-based pricing.

Who Is It For

Voice-app developers, researchers and hardware makers who want open, low-latency speech models they can study, adapt and deploy.

Use Cases

Kyutai’s open speech models suit builders of voice interfaces, robotics and accessibility tools who need low latency and inspectable models.

• Real-time conversational voice agents
• Expressive text-to-speech for media and apps
• Research on full-duplex spoken dialogue

Getting Started

Try the hosted demos first, then pull weights and inference code from the lab’s releases to prototype locally or in your stack.

Results to Expect

Developers get conversational voice fluidity that previously required closed vendors, with the freedom to modify and deploy on their own terms.

Implementation and Support

Hosted demos showcase capabilities; weights and inference code are released openly for developers to run and modify themselves.

Limitations to Keep in Mind

Open research releases mean some assembly required, and production-grade serving is still your engineering responsibility.

What Users Say

Voice-app builders are excited by full-duplex conversational quality available outside closed vendors.

Final Thoughts

Kyutai proves open science can ship cutting-edge voice AI. For builders, the practical value is simple: full-duplex conversational models you can inspect, modify and deploy without vendor permission. Expect some assembly required — but the frontier is now something you can hold in your hands rather than read about.

FAQ

Is Kyutai free to use?

Yes — released models and code are open and free; hosted endpoints may charge for production-scale usage.

Does Kyutai offer an API?

Hosted access and inference code are available for developers; production API offerings continue to expand.

How does Kyutai compare to alternatives?

Against closed voice assistants, Kyutai offers inspectable, full-duplex models at low latency; against other open speech models, Moshi’s conversational fluidity stands out.