Sakana AI: Company Profile, Products & Funding History (Unofficial)

This is an independent, third-party company profile compiled by fugu-ai.com. We have no affiliation with Sakana AI; every claim below traces to Sakana’s own published pages and dated announcements, re-checked on August 19, 2026.

Sakana AI is a Tokyo-based AI research and development company with an unusual thesis: that the future of AI looks less like one ever-bigger model and more like collective intelligence — many models, coordinated well. Founded in 2023, it went from research lab to shipping commercial products in under three years, and it built the product this site documents: Sakana Fugu, the multi-agent system delivered as one model. This profile covers who they are, what they make, how they’re funded, and where the research is headed.

The company in brief

Sakana (さかな) is Japanese for “fish,” and the company says its logo is meant to invoke a school of fish coming together — many simple individuals producing collective behavior no one of them has. That’s not just branding: nature-inspired intelligence is the company’s stated research focus, and the school-of-fish idea recurs across its work, from evolutionary optimization methods to model merging to the multi-agent orchestration that became Fugu. The product names keep the theme: fugu is a pufferfish, namazu a catfish, marlin — the deep-sea researcher — a marlin.

Sakana AI positions itself as “Building Frontier AI in Japan”: frontier-level research conducted domestically, applied to Japanese enterprise and public-sector needs, with an explicit interest in what it calls sovereignty — the ability to serve customers whose requirements constrain where models come from and where data goes.

Founders

Sakana AI’s co-founders, as named in its own materials, include Llion Jones — a co-author of “Attention Is All You Need,” the 2017 paper that introduced the Transformer architecture underlying essentially all modern language models — and David Ha, known for research on nature-inspired and evolutionary methods. A third co-founder, Ren Ito, serves as COO and appears in the company’s partnership announcements (including as advisor to MUFG under that bank’s partnership). We list only the names the official record names; this is not presented as an exhaustive founding roster.

The products

As of August 2026, Sakana AI lists five products:

• Sakana Fugu — the flagship: a multi-agent orchestration system delivered as one model, launched June 2026, offered via API, coding-agent integrations, and (since August 13, 2026) Sakana Chat. Covered in depth across this site, from pricing to the API.
• Sakana Namazu — a Japanese-specialized LLM API (launched August 3, 2026) with built-in web search and code execution, tuned for Japanese language and business contexts on an open base model.
• Sakana Marlin — the company’s first commercial product (June 15, 2026): an autonomous strategy-research agent marketed as “Your Virtual CSO,” running research sessions of up to eight hours.
• Sakana Chat — the consumer chat interface at chat.sakana.ai, which today serves Namazu (anonymous default) and Fugu (behind login), with code execution and artifact previews added August 13, 2026. Details in our chat guide.
• Sakana Translate — translation, proofreading, and Q&A features inside Sakana Chat, added July 6, 2026.

The research direction

Three threads run through Sakana’s published research. The first is collective intelligence: systems built from many coordinated parts — the TRINITY and Conductor papers behind Fugu (both ICLR 2026; our explainer), evolutionary model merging, and multi-agent work across domains. The second is efficiency over brute force — evolutionary and sample-efficient methods rather than raw compute scaling, a stance the company states explicitly in its funding announcements. The third is automating research itself: its “AI Scientist” line of work produced the first fully AI-generated paper to pass human peer review, and a paper describing that work was published in Nature (announced March 26, 2026). The same month, the company launched a dedicated lab for what it terms Recursive Self-Improvement — RSI Lab — focused, in its words, on sample-efficient self-improving AI rather than compute-hungry approaches.

A recent research note worth knowing if you follow Fugu: on August 10, 2026, Sakana published a verification showing its orchestrator (“conductor”) model can be re-trained on a different open base model — an open Gemma-family model — with performance comparable to the production conductor, which the post reveals had been trained on a Qwen-family base. The stated goal is modularity for sovereignty requirements, including future conductors built on domestically developed models.

Enterprise and public-sector work

Alongside products, Sakana AI runs an applied business working with large Japanese enterprises and government, concentrated in two publicly stated solution areas: finance and defense/intelligence.

In finance, the public record includes a comprehensive multiyear partnership with MUFG Bank (announced May 19, 2025) to develop banking-specific AI, an AI-lending project with MUFG that has moved to real-case validation, joint work with SMBC Group on multi-agent proposal-document generation, asset-consulting AI with Daiwa Securities Group (moved to a full-scale phase on August 5, 2026), and a strategic agreement with Hokkoku Financial Holdings. In defense and intelligence, the company discloses commissioned research from Japan’s defense-innovation research institute, an award at a US–Japan defense-innovation competition, work on OSINT analysis (its team placed fifth at a 2026 OSINT CTF using a Fugu-based agent), and information-analysis projects for government ministries. We describe these engagements as the company discloses them, without commentary in either direction.

Funding and partnerships: the dated record

Every entry below links a dated official announcement; amounts appear only where Sakana itself published them.

• January 16, 2024 — seed round: “$30M to develop nature-inspired AI in Japan.”
• September 4, 2024 — Series A announced.
• May 19, 2025 — multiyear comprehensive partnership with MUFG Bank.
• November 17, 2025 — Series B announced, with the company’s signature argument that current AI capital-and-compute consumption “may not be sustainable” and that Japan’s path is resource-efficient, sovereign AI.
• January 23, 2026 — strategic partnership with Google, including funding from Google and use of Google’s frontier models (the Gemini and Gemma families) in Sakana’s technology and product development.
• February 19, 2026 — investment from Salesforce Ventures.
• February 24, 2026 — strategic investment from Citi, which Citi describes as its first such investment in a Japanese company.
• February 26, 2026 — strategic partnership with Datadog: joint research, potential open-source contributions, go-to-market work.
• July 16, 2026 — partnership with NVIDIA to bring NVIDIA’s open Nemotron model family into Fugu’s orchestration pool.

Product timeline, 2026

The shipping cadence this year has been fast; all dates from Sakana’s blog (as of August 19, 2026):

• June 15 — Sakana Marlin launches as the company’s first commercial product.
• June 22 — Sakana Fugu launches: “One Model to Command Them All.”
• July 6 — Sakana Translate arrives in Sakana Chat.
• July 21 — Fugu Cyber introduced for cybersecurity reasoning.
• July 24 — a new Fugu Ultra version plus the Claude Code interface announced.
• August 3 — the Namazu API opens.
• August 13 — Sakana Chat adds Fugu (behind login) plus code execution and artifact previews.

For ongoing changes, we maintain a running, source-linked Fugu changelog.

What makes this company distinctive

Read the record above and three distinctive traits emerge. Sakana AI ships against the industry’s scaling orthodoxy — its stated bet is coordination and efficiency, not parameter count, and its flagship product is literally an argument that a well-run team of models beats any single one. It is unapologetically a Japan strategy — sovereignty, domestic enterprise, Japanese-language specialization — while publishing research and products with global reach. And it moves between paper and product unusually fast: the papers behind Fugu were published in December 2025, and the product shipped six months later. Those traits are also why this site exists: a company this fast produces facts that go stale, which is why every number here carries a date.

Start with the flagship: What is Sakana Fugu? — or meet the models directly in the playground.

fugu-ai.com is an independent, unofficial guide. We are not affiliated with, endorsed by, or sponsored by Sakana AI. This profile summarizes Sakana AI’s own published record as of August 19, 2026; for authoritative company information, consult sakana.ai.