It is getting easier to try frontier-class AI without first signing an API contract.
Japan-based Sakana AI has updated its chat service Sakana Chat and added the orchestrator model Sakana Fugu.
The company presents Sakana Chat as a place to try its models without going through an API, and the English announcement says “Try Sakana Chat for Free.”
Sakana Namazu, which is tuned for Japanese, has also been upgraded to a new generation, with support for code execution and file attachments.
The key point is that Sakana’s idea of challenging giant models with a school of smaller AIs can now be touched from a chat screen.
You can start at https://chat.sakana.ai/.
What Sakana Fugu is
Sakana Fugu is less a single giant LLM and more an orchestrator that assigns work across multiple AI models.
To users it looks like a normal chat AI, but behind the scenes it coordinates models by task and aims for performance on par with frontier models.
The company says it shines when response quality determines outcome quality, such as complex instructions or multi-step work.
On the API side there are lines such as Fugu, Fugu Ultra, and Fugu Cyber, which can also be plugged into business tools through an OpenAI-compatible endpoint.
The meaning of this Chat update is that Fugu is now easier to try in a browser before you contract and call an API.
Being able to check conversation quality first, not only paid API benchmarks, lowers the barrier to adoption.
What changed in Sakana Chat
According to the official announcement, this update has two aims.
- Expand the models you can choose (add Fugu, refresh Namazu)
- Let models roll up their sleeves and produce deliverables themselves
New capabilities include:
- Use Sakana Fugu on Sakana Chat
- Upgrade Sakana Namazu to a new generation (stronger Japanese quality and execution)
- Run Python in a sandbox
- Preview and download outputs such as HTML, Word, and slides
- Attach images, PDFs, and Office files
Until now, the chat mainly returned answers in text.
Now calculations, aggregation, and file generation can proceed in the conversation, with visual review in a right-side panel.
Uses that may fit real work
The official examples are close to everyday office tasks.
- Attach Excel and ask for quarterly totals and charts
- Pull key points from a meeting PDF and rebuild them as slides
- Paste a screenshot of an error screen and ask what went wrong
- Generate a simple game or HTML and preview it on the spot
The design lets research, calculation, and compiling materials proceed without jumping between tools.
It is also practical to use Namazu for natural Japanese documents and Fugu for more complex work.
That makes a staged path easier: prototype in chat, then connect an API to internal tools or existing systems if it fits.
Points to check before adoption
Being able to try it for free is a plus, but business use still needs checks.
- Do not confuse the free Chat trial with paid API plans (subscription or usage-based)
- Review terms and storage policy before attaching confidential files
- Understand the scope of code execution and external model routing
- Always have a human do a final check of deliverables
Because Fugu bundles multiple models, data destinations and the ability to exclude models from the pool also matter for enterprise use.
For Japanese companies, “try it in Chat first, then connect the API if it fits” is a realistic sequence.
Summary
Sakana AI’s Sakana Chat update makes Sakana Fugu easier to try, refreshes Namazu, and adds code execution and file attachments.
Being able to touch Japan-origin multi-agent AI for free from a chat screen is a meaningful shift.
Going forward, AI competition is not only about how smart one model is, but also about how models are bundled and how far they are allowed to act.
At Makoto Tejima, when planning AI chat or agent adoption, we emphasize design that includes existing-system integration, permission management, and human confirmation points.
Please feel free to consult us about business use of internal chat, OpenAI-compatible API integration, or secure agent design.
