Sakana Chat gets an update: Sakana Fugu and the new-generation Sakana Namazu are now available

Sakana AI has pushed a substantial update to Sakana Chat, its place for trying the company’s models without going through an API. The headline additions are Sakana Fugu, an orchestrator model said to match frontier-model performance, and a new-generation version of Sakana Namazu.

The bigger shift is that the chat can now execute code. That means the model can run Python in a sandbox, handle calculations and data processing, and generate files on its own instead of handing the user a set of steps to do later. The results show up in a panel on the right, where HTML is rendered in place and Word documents or slides can be reviewed immediately.

This turns Sakana Chat into more than a demo box. A user can attach an Excel file, ask for quarterly aggregation and a graph, and get the code written, run, and displayed in the same conversation. The company also says images and documents can be attached, so screenshots, PDFs, and Office files can become the starting point for questions, summaries, or analysis.

Sakana Fugu is pitched for situations where output quality depends on getting complex instructions right and moving through multi-step work cleanly. Sakana Namazu, meanwhile, has been updated from its original release to a new generation that improves both Japanese response quality and agentic execution. Sakana AI says it is especially suited to repeated daily work in Japanese, where instructions can be given directly and returned as natural documents or materials.

There’s a clear message here: Sakana wants its models to stay close to the work, not just the prompt. And once a chat can read files, run code, and redraw the output after revisions, the old “ask, then copy-paste into something else” routine starts to look a bit tired.