Sakana AI launches Namazu API for Japanese enterprise

Sakana AI has officially released the commercial API for Namazu, its specialized large language model tailored for Japanese enterprise applications. Available at console.sakana.ai, the model transitioned to a public API after being tested internally within the startup's Sakana Chat product. Built as a fine-tune of Moonshot AI's open Kimi K2.6 model rather than being trained from scratch, Namazu is designed to serve as a drop-in replacement for OpenAI APIs, requiring developers to change only two lines of code—the base URL and the API key—to integrate it into existing workflows.

The fine-tuning process leverages Sakana AI's proprietary datasets to adapt Kimi K2.6 specifically for Japanese linguistic and cultural contexts. This targeted optimization yields significant performance gains on regional benchmarks. On the FairPoliticsQA evaluation, Namazu boosted the base model's score from 34.10% to 56.30%, while also outperforming Kimi K2.6 on JFBench and translation tasks. It retains the core strengths of its base model, including support for text, image, and video inputs, thinking and non-thinking modes, and long-horizon coding capabilities, while reducing unnecessary refusals and output bias.

For practitioners building agentic workflows, Namazu simplifies development by integrating built-in web search and code execution directly into the model, eliminating the need for external scaffolding. It addresses a critical market gap for Japanese enterprises that find frontier models too expensive but struggle with the quality and data-handling limitations of raw open-weight models. By offering a middle path, the Tokyo-based startup—founded in 2023 by David Ha and Llion Jones—provides a cost-effective, culturally fluent alternative for local business applications.