Sakana AI debuts Marlin, an AI analyst that works 8 hours straight

Most AI search tools are built around instant gratification: you type a prompt, wait ten seconds, and get three tidy bullet points. But if you actually need to build a board-level strategy deck, that gets you nowhere.

Tokyo-based Sakana AI launched beta testing for its very first commercial product: Sakana Marlin. The company calls it a "virtual Chief Strategy Officer," and it is built to do the exact opposite of quick search—it deliberately takes its time.

Once you hand Marlin a complex strategic prompt, human intervention is over. The system sets off on an autonomous research marathon lasting nearly eight hours, eventually returning a structured presentation deck and a comprehensive report running dozens of pages long.

What an 8-hour reasoning agent looks like

In corporate strategy, producing deep research usually takes a team of junior consultants several weeks. Sakana AI wants Marlin to do that legwork in an afternoon.

To prove the concept, the startup shared sample outputs. One is a 61-page report dissecting how Trump's trade policies affect Japanese supply chains, breaking down 16 structural risk factors across semiconductors and pharma. Another is a 78-page deep dive into generative AI ROI across Japanese banking.

These aren't just massive text dumps. They include executive summaries, cross-sector scenarios, and structured references.

Behind this is the concept of test-time compute. Sakana points to Jevons Paradox: as model inference becomes more efficient, the winning play isn't doing tasks faster, but pouring massive compute into reasoning longer on harder problems.

Monte Carlo trees meet automated science

Letting an LLM spin its wheels for eight hours usually leads to hallucinations. Marlin stays on track by leaning on two of Sakana's core research breakthroughs.

The first is AB-MCTS (Adaptive Branching Monte Carlo Tree Search), a technique highlighted at NeurIPS 2025. Instead of following a single train of thought, Marlin treats reasoning like a game tree, testing hypotheses across hundreds or thousands of model calls, pruning dead ends, and swapping between specialized AI models when needed.

The second pillar comes directly from The AI Scientist, Sakana's automated research framework featured in Nature. Marlin applies that same scientific cycle—hypothesis, evidence hunting, contradiction checks, and structured writeup—to corporate problems.

Sakana is currently recruiting corporate strategy teams, consultancies, financial institutions, and think tanks for a closed beta, which is free during the testing phase.

Why it matters

While the rest of the AI industry is busy cramming faster chatbots into consumer apps, Sakana AI is aiming straight at high-margin corporate consulting. It's a very smart bet: the business world doesn't pay premium prices for quick answers, it pays for exhaustive rigor.

For strategists, market analysts, and consultants, tools like Marlin could flip the daily grind upside down. Instead of spending 30 hours digging through public filings and drafting slides, analysts can shift their attention to framing the research question and stress-testing the output.

That said, reviewing a 78-page AI-generated report creates its own headaches. Fact-checking an eight-hour autonomous reasoning chain requires deep domain expertise, otherwise teams risk accepting polished nonsense. If Sakana can nail verification and source transparency, Marlin could easily become the default starting point for corporate decision-making.