Sakana AI launched Marlin on June 15, 2026. It is the company's first commercial product, and it is aimed squarely at strategy teams, finance departments, and consulting firms that spend weeks on research before a big decision.
The product works like this. You give it one research topic. It then runs on its own for up to eight hours, forming hypotheses, browsing sources, and checking its own findings. At the end, you receive a report of up to 100 pages with citations, plus a presentation slide deck ready to share with leadership. At a press hands-on session, reports ran between 60 and 100 pages and cited 60 to 80 sources.
This is very different from the deep research tools most professionals have already tried. Tools from OpenAI, Google, and Perplexity return results in minutes. Marlin deliberately takes hours. The trade-off is explicit: you wait longer, and in return you get a more structured, more deeply tested output. The question is whether the extra depth is worth the wait and the cost, and for a genuinely high-stakes question, it often will be.
The pricing reflects that this is not a casual tool. Pay-as-you-go starts at 100 credits per run, at 98 yen each. The Pro plan runs 150,000 yen per month, which is around $1,000 at current exchange rates, and includes 2,000 credits. A Team plan at 400,000 yen per month adds more credits and multi-user access. Enterprise pricing is custom.
One detail that matters for any professional handling sensitive information: Sakana has stated that neither the company nor any of its AI service providers will use your research topics or inputs to train their models, unless you explicitly agree. This is not unique in the industry, but it is the right baseline for anything involving unreleased strategy, competitive intelligence, or financial data.
Sakana is not a small startup taking a long shot. The company has raised around $335 million across its Series A and Series B rounds, reaching a valuation of about $2.65 billion. Its investors include MUFG, Japan's largest bank, as well as Citigroup, Salesforce, and Google. Citi's investment, announced in February 2026, was its first strategic investment in a Japanese company. MUFG has a separate multi-year partnership with Sakana to build AI tools specifically for banking operations.
The company was founded in July 2023 by former Google researchers and has been focused on applying AI to tasks that require real depth, not just fast replies. Marlin is built on research Sakana published about a reasoning method called Adaptive Branching Tree Search. The short version: instead of just generating many possible answers and picking the best one, the system decides at each step whether to explore a new direction or dig deeper into a promising one. This makes it better suited to long, complex research tasks than standard AI tools.
Who should pay attention? Anyone whose job involves commissioning or producing strategy documents, due diligence reports, market assessments, or competitive briefings. A run of Marlin does not replace judgment or domain expertise. A human still needs to review the output before any decision. But if the bottleneck in your process is the weeks it takes to gather and synthesize information into a presentable format, this kind of tool directly addresses that problem.
For now, Marlin is only available as a cloud service. You cannot install it on your own servers. That is a constraint worth noting for industries with strict data residency rules, though the no-training data policy provides some reassurance. The company is clearly positioning itself for regulated industries: finance, consulting, and now, according to public reports, defense.