이미지: X — 모델·오픈소스
Summary
- Sakana AI has added new-generation models Namazu and Fugu to its Sakana Chat service
- The models specialize in Japanese culture and support vibe coding, letting users create games and apps from text descriptions alone
- Attaching an Excel file triggers automatic chart creation and report generation
- 발표 주체
- Sakana AI (공식 X 계정 @SakanaAILabs)
- 발표일
- 2026-08-13
- 신규 모델
- Namazu(메기), Fugu(복어)
- 탑재 서비스
- Sakana Chat
- 핵심 기능
- 바이브 코딩 — 말로 설명하면 게임·앱 생성
- 예시 활용
- 한자 드릴, 금붕어 잡기 게임, 엑셀 데이터 분석·자료화
- 특화 영역
- 일본어·일본 문화, 일본 비즈니스 현장
Two new models named after fish
Tokyo-based Sakana AI has added two new-generation models to its Sakana Chat service. Their names: Namazu (catfish) and Fugu (pufferfish). Since the company name "Sakana" means fish in Japanese, the model names follow the same theme.
Sakana AI described the new models as being "deeply versed in Japanese culture," and said they implement vibe coding — the recently popular approach where AI builds games or apps from spoken descriptions rather than hand-written code. According to the company, ideas ranging from kanji practice worksheets to a goldfish-catching game can be turned into working results.
What's this about
A day earlier, on August 12, Sakana AI had already teased a robotic fish project called "MECHA FISH" with a preview image. The image carried the text "in development" alongside the subtitle "THE FUTURE OF AUTONOMOUS SHOALING." This connects to the company's expansion of RSI Lab on August 10, when it positioned physical AI as the next frontier of recursive self-improvement. The Namazu and Fugu announcement, by contrast, is on the software side — a generational update to the chatbot models.
Vibe coding is a concept that has spread this year beyond the developer community to general users. It lets people describe requirements in natural language, without programming knowledge, and have AI immediately generate a working result — a trend led by coding tools like Cursor and Lovable. Sakana AI appears to be differentiating itself by bringing this approach into chatbot form and tailoring it specifically for Japanese-language users.
What was shown
The released sample image shows a screen where an Excel file named "受注詳細_サンプルデータ.xlsx" is attached. Alongside the message that a spreadsheet turns into insight, it demonstrates the Namazu model handling everything from data analysis to document creation.
How to use it
Sakana AI outlined two specific use cases in the announcement.
The first is educational content creation. Speaking an idea into Sakana Chat produces a result that can be used right away. For example, during summer vacation, a parent and child could build and solve arithmetic drills together in a browser, or create a kanji-learning game on the spot.
The second is business data analysis. Simply attaching an Excel file is said to be enough to generate charts and a draft report. Given the emphasis on Japanese-language processing, this appears geared toward document work at Japanese companies.
Details on how to access the service, pricing plans, and supported countries were not clearly specified in this announcement.
The announcement at a glance
| Item | Details |
|---|---|
| New models | Namazu (catfish), Fugu (pufferfish) |
| Service | Sakana Chat |
| Key feature | Vibe coding (text-to-game/app generation) |
| Example use cases | Kanji drills, goldfish-catching game, Excel analysis and reporting |
| Target users | Japanese-language users, Japanese business settings |
So what changes
For users looking for a Japanese-specialized AI chatbot, there's now an added option to create learning games or work documents on the spot, without any programming knowledge. In particular, the workflow of pasting in Excel data and receiving analysis results directly could translate into real time savings for professionals who spend hours on repetitive reporting tasks. The fact that Sakana AI rolled out both a robotic fish project and a chatbot model upgrade in the same week suggests the company is trying to expand its presence simultaneously in both software and physical AI.
