Kyutai

Team

non-profit

Verified

https://kyutai.org/

kyutai_labs

kyutai-labs

Activity Feed

AI & ML interests

Our mission is to build and democratize artificial general intelligence through open science

Recent Activity

ameroyer updated a Space 1 day ago

kyutai/README

gabrielatkyutail updated a Space 1 day ago

kyutai/README

gabrielatkyutail published a Space 1 day ago

kyutai/README

View all activity

Papers

The FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation

Understanding Data Temporality Impact on Large Language Models Pre-training

View all Papers

kyutai 's collections 12

Interactivity Alignment

Full-duplex speech models post-trained with reinforcement learning for improved conversational interactivity.

Multi-Faceted Interactivity Alignment in Full-Duplex Speech Models

Paper • 2606.11167 • Published Jun 9 • 5
kyutai/moshika-rl-seamless

Audio-to-Audio • 8B • Updated Jun 11 • 218 • 27
kyutai/personaplex-rl-seamless

Audio-to-Audio • 8B • Updated Jun 11 • 2.55k • 19
kyutai/interactivity-alignment-samples

Viewer • Updated Jun 10 • 5.56k • 1.05k • 8

Hibiki-Zero

Streaming speech translation without the need for word-level alignments

Running

13

Hibiki Zero Samples

🏆

13

Demo samples of the speech translation model Hibiki-Zero.
Simultaneous Speech-to-Speech Translation Without Aligned Data

Paper • 2602.11072 • Published Feb 11 • 2
kyutai/Audio-NTREX-4L

Viewer • Updated Feb 12 • 3.6k • 578 • 6
kyutai/hibiki-zero-3b-pytorch-bf16

Audio-to-Audio • Updated Feb 12 • 4.7k • 56

CASA

CASA: Cross-Attention over Self-Attention for Efficient Vision-Language Fusion on long-context streaming inputs

Running

Agents

4

CASA Gallery

🏠

4

Video Gallery for CASA: Cross-Attention over Self-Attention
CASA: Cross-Attention via Self-Attention for Efficient Vision-Language Fusion

Paper • 2512.19535 • Published Dec 22, 2025 • 13
kyutai/CASA-Helium1-VL-2B

Image-Text-to-Text • 3B • Updated Mar 9 • 13 • 9
kyutai/CASA-Qwen2_5-VL-3B

Image-Text-to-Text • 4B • Updated Dec 23, 2025 • 9 • 3

Text-To-Speech

https://kyutai.org/next/tts

kyutai/pocket-tts

Updated May 4 • 1.56k • 734
kyutai/pocket-tts-without-voice-cloning

Updated May 4 • 5.86k • 29
kyutai/tts-1.6b-en_fr

Text-to-Speech • Updated Sep 11, 2025 • 30.1k • 378
kyutai/tts-voices

Updated Mar 9 • 166

Helium 1

Helium 1: a modular and multilingual LLM

kyutai/helium-1-2b

Text Generation • 2B • Updated Apr 30, 2025 • 23.1k • 56
kyutai/helium-1-2b-books

Text Generation • 2B • Updated Apr 30, 2025 • 35 • 2
kyutai/helium-1-2b-hum

Text Generation • 2B • Updated Apr 30, 2025 • 41 • 1
kyutai/helium-1-2b-life

Text Generation • 2B • Updated Apr 30, 2025 • 50 • 2

Hibiki fr-en

Hibiki is a model for streaming speech translation , which can run on device! See https://github.com/kyutai-labs/hibiki.

Running

54

Hibiki Samples

🤗

54

Translate speech in real-time with high fidelity
High-Fidelity Simultaneous Speech-To-Speech Translation

Paper • 2502.03382 • Published Feb 5, 2025 • 10
kyutai/hibiki-1b-mlx-bf16

Translation • Updated Feb 6, 2025 • 36 • 31
kyutai/hibiki-2b-mlx-bf16

Translation • Updated Feb 6, 2025 • 8 • 23

MoshiRAG Release

Candle & PyTorch model checkpoints released as part of the MoshiRAG release from Kyutai. Run inference via: https://github.com/kyutai-labs/moshi-rag

MoshiRAG: Asynchronous Knowledge Retrieval for Full-Duplex Speech Language Models

Paper • 2604.12928 • Published Apr 14 • 2
kyutai/moshika-rag-pytorch-bf16

Audio-to-Audio • 8B • Updated Apr 17 • 2.11k • 10
kyutai/moshika-rag-candle-bf16

Audio-to-Audio • 8B • Updated Apr 17 • 186 • 12

Kairos

Temporal pretraining checkpoints and KairosQA evaluation dataset

kyutai/KairosQA

Viewer • Updated May 26 • 7.17k • 209 • 1
Understanding Data Temporality Impact on Large Language Models Pre-training

Paper • 2605.22769 • Published May 21 • 8
kyutai/Sequential_Helium_6B

Text Generation • 6B • Updated May 26 • 137 • 1

ARC-Encoders

Pretrained ARC-Encoders and a fine-tuning dataset: context compression for unmodified LLMs.

ARC-Encoder: learning compressed text representations for large language models

Paper • 2510.20535 • Published Oct 23, 2025 • 10
kyutai/ARC8_Encoder_Llama

Feature Extraction • Updated Nov 5, 2025 • 20 • 3
kyutai/ARC_finetuning

Preview • Updated Oct 24, 2025 • 55 • 1
kyutai/ARC8_Encoder_multi

Feature Extraction • Updated Nov 5, 2025 • 6 • 7

Speech-To-Text

https://kyutai.org/next/stt

kyutai/stt-2.6b-en

Automatic Speech Recognition • 3B • Updated Jun 26, 2025 • 124
kyutai/stt-1b-en_fr

Automatic Speech Recognition • 1.0B • Updated Nov 18, 2025 • 132
kyutai/stt-1b-en_fr-mlx

Automatic Speech Recognition • Updated Jun 19, 2025 • 6
kyutai/stt-2.6b-en-mlx

Automatic Speech Recognition • Updated Jun 19, 2025 • 9

MoshiVis v0.1

MoshiVis is a Vision Speech Model built as a perceptually-augmented version of Moshi v0.1 for conversing about image inputs

Vision-Speech Models: Teaching Speech Models to Converse about Images

Paper • 2503.15633 • Published Mar 19, 2025 • 3
kyutai/Babillage

Viewer • Updated Mar 21, 2025 • 465k • 554 • 14
kyutai/moshika-vis-pytorch-bf16

Updated Jun 18, 2025 • 59
kyutai/moshika-vis-candle-bf16

Updated Mar 18, 2025 • 2

Moshi v0.1 Release

MLX, Candle & PyTorch model checkpoints released as part of the Moshi release from Kyutai. Run inference via: https://github.com/kyutai-labs/moshi

Moshi: a speech-text foundation model for real-time dialogue

Paper • 2410.00037 • Published Sep 17, 2024 • 18
kyutai/moshiko-pytorch-bf16

Updated Sep 18, 2024 • 175k • 244
kyutai/moshika-pytorch-bf16

Updated Sep 18, 2024 • 13.3k • 64
kyutai/mimi

Feature Extraction • 96.2M • Updated Jul 2, 2025 • 491k • • 311

Interactivity Alignment

Full-duplex speech models post-trained with reinforcement learning for improved conversational interactivity.

Multi-Faceted Interactivity Alignment in Full-Duplex Speech Models

Paper • 2606.11167 • Published Jun 9 • 5
kyutai/moshika-rl-seamless

Audio-to-Audio • 8B • Updated Jun 11 • 218 • 27
kyutai/personaplex-rl-seamless

Audio-to-Audio • 8B • Updated Jun 11 • 2.55k • 19
kyutai/interactivity-alignment-samples

Viewer • Updated Jun 10 • 5.56k • 1.05k • 8

MoshiRAG Release

Candle & PyTorch model checkpoints released as part of the MoshiRAG release from Kyutai. Run inference via: https://github.com/kyutai-labs/moshi-rag

MoshiRAG: Asynchronous Knowledge Retrieval for Full-Duplex Speech Language Models

Paper • 2604.12928 • Published Apr 14 • 2
kyutai/moshika-rag-pytorch-bf16

Audio-to-Audio • 8B • Updated Apr 17 • 2.11k • 10
kyutai/moshika-rag-candle-bf16

Audio-to-Audio • 8B • Updated Apr 17 • 186 • 12

Hibiki-Zero

Streaming speech translation without the need for word-level alignments

Running

13

Hibiki Zero Samples

🏆

13

Demo samples of the speech translation model Hibiki-Zero.
Simultaneous Speech-to-Speech Translation Without Aligned Data

Paper • 2602.11072 • Published Feb 11 • 2
kyutai/Audio-NTREX-4L

Viewer • Updated Feb 12 • 3.6k • 578 • 6
kyutai/hibiki-zero-3b-pytorch-bf16

Audio-to-Audio • Updated Feb 12 • 4.7k • 56

Kairos

Temporal pretraining checkpoints and KairosQA evaluation dataset

kyutai/KairosQA

Viewer • Updated May 26 • 7.17k • 209 • 1
Understanding Data Temporality Impact on Large Language Models Pre-training

Paper • 2605.22769 • Published May 21 • 8
kyutai/Sequential_Helium_6B

Text Generation • 6B • Updated May 26 • 137 • 1

CASA

CASA: Cross-Attention over Self-Attention for Efficient Vision-Language Fusion on long-context streaming inputs

Running

Agents

4

CASA Gallery

🏠

4

Video Gallery for CASA: Cross-Attention over Self-Attention
CASA: Cross-Attention via Self-Attention for Efficient Vision-Language Fusion

Paper • 2512.19535 • Published Dec 22, 2025 • 13
kyutai/CASA-Helium1-VL-2B

Image-Text-to-Text • 3B • Updated Mar 9 • 13 • 9
kyutai/CASA-Qwen2_5-VL-3B

Image-Text-to-Text • 4B • Updated Dec 23, 2025 • 9 • 3

ARC-Encoders

Pretrained ARC-Encoders and a fine-tuning dataset: context compression for unmodified LLMs.

ARC-Encoder: learning compressed text representations for large language models

Paper • 2510.20535 • Published Oct 23, 2025 • 10
kyutai/ARC8_Encoder_Llama

Feature Extraction • Updated Nov 5, 2025 • 20 • 3
kyutai/ARC_finetuning

Preview • Updated Oct 24, 2025 • 55 • 1
kyutai/ARC8_Encoder_multi

Feature Extraction • Updated Nov 5, 2025 • 6 • 7

Text-To-Speech

https://kyutai.org/next/tts

kyutai/pocket-tts

Updated May 4 • 1.56k • 734
kyutai/pocket-tts-without-voice-cloning

Updated May 4 • 5.86k • 29
kyutai/tts-1.6b-en_fr

Text-to-Speech • Updated Sep 11, 2025 • 30.1k • 378
kyutai/tts-voices

Updated Mar 9 • 166

Speech-To-Text

https://kyutai.org/next/stt

kyutai/stt-2.6b-en

Automatic Speech Recognition • 3B • Updated Jun 26, 2025 • 124
kyutai/stt-1b-en_fr

Automatic Speech Recognition • 1.0B • Updated Nov 18, 2025 • 132
kyutai/stt-1b-en_fr-mlx

Automatic Speech Recognition • Updated Jun 19, 2025 • 6
kyutai/stt-2.6b-en-mlx

Automatic Speech Recognition • Updated Jun 19, 2025 • 9

Helium 1

Helium 1: a modular and multilingual LLM

kyutai/helium-1-2b

Text Generation • 2B • Updated Apr 30, 2025 • 23.1k • 56
kyutai/helium-1-2b-books

Text Generation • 2B • Updated Apr 30, 2025 • 35 • 2
kyutai/helium-1-2b-hum

Text Generation • 2B • Updated Apr 30, 2025 • 41 • 1
kyutai/helium-1-2b-life

Text Generation • 2B • Updated Apr 30, 2025 • 50 • 2

MoshiVis v0.1

MoshiVis is a Vision Speech Model built as a perceptually-augmented version of Moshi v0.1 for conversing about image inputs

Vision-Speech Models: Teaching Speech Models to Converse about Images

Paper • 2503.15633 • Published Mar 19, 2025 • 3
kyutai/Babillage

Viewer • Updated Mar 21, 2025 • 465k • 554 • 14
kyutai/moshika-vis-pytorch-bf16

Updated Jun 18, 2025 • 59
kyutai/moshika-vis-candle-bf16

Updated Mar 18, 2025 • 2

Hibiki fr-en

Hibiki is a model for streaming speech translation , which can run on device! See https://github.com/kyutai-labs/hibiki.

Running

54

Hibiki Samples

🤗

54

Translate speech in real-time with high fidelity
High-Fidelity Simultaneous Speech-To-Speech Translation

Paper • 2502.03382 • Published Feb 5, 2025 • 10
kyutai/hibiki-1b-mlx-bf16

Translation • Updated Feb 6, 2025 • 36 • 31
kyutai/hibiki-2b-mlx-bf16

Translation • Updated Feb 6, 2025 • 8 • 23

Moshi v0.1 Release

MLX, Candle & PyTorch model checkpoints released as part of the Moshi release from Kyutai. Run inference via: https://github.com/kyutai-labs/moshi

Moshi: a speech-text foundation model for real-time dialogue

Paper • 2410.00037 • Published Sep 17, 2024 • 18
kyutai/moshiko-pytorch-bf16

Updated Sep 18, 2024 • 175k • 244
kyutai/moshika-pytorch-bf16

Updated Sep 18, 2024 • 13.3k • 64
kyutai/mimi

Feature Extraction • 96.2M • Updated Jul 2, 2025 • 491k • • 311

AI & ML interests

Recent Activity

Papers

Team members 20

kyutai 's collections 12

Hibiki Zero Samples

CASA Gallery

Hibiki Samples

Hibiki Zero Samples

CASA Gallery

Hibiki Samples