Building Frontier AI in Japan
One model that orchestrates the world's best LLMs
The best LLM for Japanese, at a reasonable price
A showcase of our post-training process, with web search and code execution built in.
The ultimate solution for strategic decision-making
An autonomous research agent that reasons for hours and returns analyst-grade strategy reports for M&A consultants, strategy departments, and think tanks.
Best-in-class for conversational and polite Japanese
Built on Namazu. Translate and proofread across Japanese, English, and Chinese, with the right register from casual to formal.
Our Applied team works as an extension of yours: scoping the use case with your experts, then building and evaluating the system until it holds up in the field. That is how we work with MUFG Bank, SMBC Group and Daiwa Securities Group in finance, and with the Ministry of Defense, ATLA and the Ministry of Internal Affairs and Communications in defense and intelligence.
Selected case studies (Japanese)
Built under a multiyear partnership, now being validated on live loan cases.
A jointly developed application in which multiple AI agents draft client proposals.
The joint project enters its full-scale phase with AI support for wealth management.
Technology that maps the information space and counters false and misleading information, built within an MIC program.
Contracted to study and demonstrate the AI capabilities that integrated analysis requires.
Research for the Defense Innovation Science and Technology Institute on combining AI technologies for reporting, information integration and resource allocation.
Our work on evolution, collective intelligence, and self-improving systems appears at ICLR, NeurIPS, and in Nature, and ships in our products.
Selected work
Towards fully automated AI research, published in Nature.
An evolved coordinator that hands each step of a task to the right model. The first paper behind Sakana Fugu.
Learning to orchestrate agents in natural language. The second paper behind Sakana Fugu.
Inference-time scaling and collective intelligence for frontier AI.
AI that improves itself by rewriting its own code.
A new neural architecture that uses time to think.
Reshaping sparsity to fit the GPU makes LLMs faster and lighter. With NVIDIA.
Evolving new algorithms with LLMs, orders of magnitude more efficiently.
Can AI agents be creative when nobody tells them what to create?
Towards collective intelligence for the physical world.
We are hiring across Research, Product, Corporate and Enterprise Solutions (Applied Team).