Sakana Marlin is the Tokyo-based AI lab's first commercial product, and it is not a chatbot. It is an autonomous research agent designed to sit down with a strategic question and not surface for up to eight hours, emerging with a structured slide deck and a multi-dozen-page report. No human in the loop. No follow-up prompts required.

Sakana AI has launched a closed beta for Marlin, marking a notable evolution from its previous consumer-oriented Sakana Chat service. Marlin is designed to emulate the work of a CSO and an entire research team, compressing several weeks of strategic analysis into a single, unattended session lasting up to eight hours. The announcement also confirms a full self-serve commercial launch with tiered plans ranging from free pay-per-use up through Pro, Team, and Enterprise.

Years of research, one product

Marlin is not a wrapper around a single frontier model. It is built upon two core research developments Sakana has refined over the past two years. The first is Adaptive Branching Monte Carlo Tree Search (AB-MCTS), which conceptualizes reasoning as a tree-search problem, allowing the AI to autonomously determine which hypotheses to pursue, discard, or escalate to different models. This technique was recognized with a spotlight award at NeurIPS 2025.

AB-MCTS is worth understanding in more detail. At its core, it reflects a philosophical shift in how we think about LLM reasoning. Rather than treating generation as a flat, linear process, Sakana's approach reframes inference as a strategic exploration through a search tree, navigating between depth (refinement of existing ideas), width (generation of new hypotheses), and even model selection itself. Think of it as the AI equivalent of a research team that knows when to go deeper on a promising lead versus when to pivot and explore a new angle entirely.