Source-led article
Sakana AI Launches Marlin, an Enterprise Agent for In-Depth Research Reports
Sakana AI, a company based in Tokyo, has released its first commercial product, Sakana Marlin. Positioned as a “Virtual CSO” (Chief Strategy Officer), Marlin is an autonomous B2B research agent built specifically for enterprise use. Unlike chatbots that provide instant answers, Marlin operates autonomously for up to eight hours on a single research topic, delivering extensive reports and accompanying presentation slide decks.
Marlin is designed to compress weeks of strategic work into mere hours. It functions by taking a research topic or question, then independently planning hypotheses, browsing various sources, and verifying its findings. The output is structured to aid decision-makers, with reports often ranging from 60 to 100 pages, citing 60 to 80 sources, and including main body content, references, and appendices. Presentation slides are generated using AI image generation capabilities.
Key facts
| Feature | Description |
|---|---|
| Product Name | Sakana Marlin |
| Function | Autonomous enterprise research agent |
| Output | Up to 100-page reports, presentation slides |
| Core Technology | Adaptive Branching Monte Carlo Tree Search (AB-MCTS) |
| Autonomy | Up to 8 hours per task |
The Technology Behind Marlin
The core of Sakana Marlin is Adaptive Branching Monte Carlo Tree Search (AB-MCTS), a technology derived from Sakana AI’s research paper, “Wider or Deeper? Scaling LLM Inference-Time Compute with Adaptive Branching Tree Search.” AB-MCTS frames reasoning as a tree-search problem where the algorithm continually decides whether to broaden its search for new candidate answers or deepen its exploration of promising existing ones. This adaptive approach contrasts with standard methods that typically only go wider, hoping for a correct answer.
Marlin also integrates workflow automation from Sakana’s AI Scientist project, which previously demonstrated autonomous scientific discovery and was featured in Nature. This combination allows Marlin to apply adaptive search to complex, long-horizon research tasks, enhancing its ability to conduct in-depth analysis. An interactive demo, `marlin-abmcts-demo.html`, illustrates this “wider or deeper” decision-making process live, showing how the search tree grows and highlights the best path based on scores.
Beta Testing and Enterprise Adoption
Sakana AI refined Marlin through a closed beta program in April 2026, where approximately 300 professionals tested the agent on real-world tasks. These tasks included strategy formulation, market research, risk analysis, and competitive analysis. The company has also secured a strategic investment from Citigroup and partnered with MUFG, indicating growing confidence in its enterprise solution.
While Marlin prioritizes depth over speed, it offers a trade-off: longer processing times (hours rather than minutes) for higher-quality, more thoroughly vetted outputs. Users can cancel a run at any time, though credits are still consumed. Sakana offers various pricing tiers, including pay-as-you-go, Pro, Team, and custom Enterprise options, with pay-as-you-go starting at 100 credits per run at ¥98 per credit.
Open-Source Contribution
For developers and researchers interested in its underlying mechanics, Sakana AI has open-sourced AB-MCTS as TreeQuest under the Apache 2.0 license. This allows users to install the algorithm, define a generation function, and run a fixed search budget, enabling experimentation with the core technology. The open-source version also supports multi-LLM search and checkpointing for managing long-running sessions.
Marlin is designed for high-stakes questions where research is a bottleneck, providing concrete examples of its utility in strategy and market analysis. It represents a significant step towards more autonomous and in-depth AI-driven research capabilities for businesses.