Japanese AI research lab · Tokyo

Sovereign AI for Japan,
built in the open.

We build Japanese-first models, advance the tools behind them, and turn our research into products people use.

Explore our research Explore our products

From data to deployment. Hover an area to see our work in it.

Officially listed as a 国産基盤モデル開発事業者→ Japanese Foundation Model Developer under METI & NEDO's GENIAC program.

Open weights · open data

Models anyone can run

The Shisa V2 and V2.1 bilingual model families are published on Hugging Face, alongside the training mixtures that built them. Our open-source models have been downloaded millions of times.

Hugging Facehuggingface.co/shisa-ai

Across the full stack

Leading-edge tools, in the open

We publish research, build and maintain open-source tools, and contribute upstream. Our work spans model training and evaluation, inference, coding agents, memory, and security, and more.

GitHubgithub.com/shisa-ai

Platform

Build with the best of open AI.

Models, developer tools, and Japanese speech, ready for your next product.

Power your coding workflow.

Jouzu is a coding harness that puts a team of leading open coding models on one codebase: workers, a judge and a verifier, each on the model that suits the job.

GLM 5.3
DeepSeek v4 Pro
Kimi K3
Qwen 3.8 Max
Gemini 3 Pro
Shisa V2.1 70B

npm install -g jouzu

Explore Jouzu

Shisa Platform

Leading models. One API.

Our own Shisa models next to leading partner models, behind one OpenAI-compatible API, with pricing in yen or dollars.

Partner models

Kimi K3
GLM 5.3
Qwen 3.8 Max
Qwen 3.8 Flash
Gemini 3 Pro

Shisa models

Shisa 2.1 Flash
Shisa 2.1 Pro

Browse models

Japanese speech, understood and spoken.

5.0%: character error rate on the SPREDS Japanese benchmark
~39 ms: engine processing latency
JP: all processing on servers in Japan, Kansai-ben included

Real-time ASR, measured in Shisa's test environment, Sept 2026.

Hear it in action Text to speech

Translation

Do business in Japanese like a native.

Whether you’re presenting on stage, meeting in person, or joining a video call, Chotto helps everyone follow the conversation. Real-time translation in Japanese, English, and Chinese, on the big screen, your desktop, or your phone.