Sakana AI Marlin Review: The 8-Hour Research Agent

Most AI tools feel like they’re begging for your attention. They want you to prompt them, poke them, and refine their output until you’ve basically done the work yourself. Sakana AI’s Marlin, launched today from the Tokyo-based research lab, takes a different approach. It’s the first tool that feels like it actually wants to be left alone.

While the rest of the industry is obsessed with reducing latency to milliseconds, Sakana AI co-founders Llion Jones and David Ha are betting on the opposite: the value of slow thinking. Marlin isn’t a chatbot you talk to; it’s an autonomous analyst you brief once and then ignore for the next eight hours. By the time you check back, it has produced a 100-page strategy report that would have cost a consulting firm six figures and three weeks of billable hours.

What is Sakana AI Marlin?

Marlin is a commercial evolution of the lab’s widely cited AI Scientist project. At its core, it is an autonomous research agent designed to function as a ‘Virtual Chief Strategy Officer.’ You give it a high-level objective—say, an analysis of the semiconductor supply chain in Southeast Asia over the next five years—and it goes to work.

It doesn’t just scrape the web. It uses a technique called Adaptive Branching Monte Carlo Tree Search (AB-MCTS) to navigate the web like a human researcher would. It forms a hypothesis, searches for evidence, finds a contradiction, and then ‘branches’ its search to resolve that conflict. This is a significant step beyond the basic retrieval-augmented generation (RAG) we see in most chatgpt-alternatives. Marlin is actually doing the legwork, making thousands of calls to various models and APIs to verify every claim it makes.

How does Marlin differ from Perplexity?

The comparison to Perplexity is inevitable, but it’s a category error. Perplexity is an answer engine; it’s built to tell you what something is. When you use is-perplexity-pro-worth-it, you’re looking for a faster way to find facts that already exist.

Marlin is built to figure out what to do. It thrives on open-ended strategy questions where the data is messy and the ‘right’ answer depends on synthesizing multiple conflicting sources. Perplexity might take 20 minutes to give you a dense, citation-heavy summary. Marlin takes 8 hours to give you a roadmap. It’s the difference between a research librarian and a senior analyst.

For a business owner deciding between the two, it comes down to the scale of the problem. If you’re looking for the best ai-writing-tools, you don’t need Marlin. If you’re looking to enter a new market or conduct due diligence on a $50 million acquisition, you do.

The tech under the hood: Nature-inspired reasoning

Sakana AI has always been the ‘weird’ lab in the best possible way. While San Francisco labs focus on scaling compute, the Tokyo team looks at nature. The name ‘Sakana’ means fish, and their research often mimics collective intelligence and evolutionary biology.

Marlin uses a ‘model merging’ architecture where multiple specialized models cooperate. Instead of one giant, expensive model trying to do everything, Marlin orchestrates a school of smaller, efficient models. This allows it to run for hours without the astronomical compute costs usually associated with long-context reasoning. This efficiency is why it can maintain a high resolution of detail over a 100-page report without the ‘forgetfulness’ or hallucinations that plague other large language models.

This launch comes at a pivot point for the industry. Today’s news that Salesforce is acquiring Fin (formerly Intercom) for $3.6 billion proves that the era of the ‘customer bot’ is ending and the era of the ‘agent’ is starting. Salesforce wants to dominate the customer service agent space; Sakana AI is staking its claim on the high-end strategy agent.

Is Sakana AI Marlin worth the enterprise price?

Marlin isn’t cheap, and it isn’t meant to be. It’s priced for the ‘Pro’ and ‘Enterprise’ crowd who view time as their most expensive resource. The value proposition is simple: if Marlin can replace 60% of the work done by a junior strategy team, the ROI is immediate.

In our testing, the output is remarkably polished. The reports aren’t just walls of text; they include structured data, competitive matrices, and executive-ready slide decks. It’s the kind of work that usually requires a human to sit in a room for a week with twelve tabs open. Marlin does it in the background while you’re in meetings.

However, there is a learning curve. You have to be precise with your initial brief. If you give it a vague prompt, you’ll get 100 pages of vague ‘strategy.’ To get the most out of it, you need to treat it like a high-performing employee—give it clear constraints, specific goals, and a defined scope. For those looking for the best-ai-tools-for-small-business, Marlin might be overkill, but for a growing firm, it’s a force multiplier.

Final Verdict

Sakana AI has successfully moved from a ‘lab to watch’ to a ‘company to buy from.’ Marlin is a sophisticated, specialized tool that recognizes a fundamental truth about high-level work: it takes time to think. By productizing that thinking time, Sakana has created something that feels truly new in a market crowded with wrappers and clones.

Marlin doesn’t just answer your questions. It does your homework. And in 2026, that is the only thing that matters.