Sakana AI, the Tokyo-based lab known for its work on automated scientific discovery, launched its first commercial product on June 15. Marlin is an autonomous research agent that accepts a topic, runs without human input for roughly eight hours, and returns executive summary slides alongside detailed written reports that can reach one hundred pages.
The pitch to buyers is a fully hands-off research loop. Once parameters are set, Marlin develops its own hypotheses, gathers information across sources, validates findings, maps causal relationships between factors, and organizes outputs into strategic options formatted for leadership review. The company says its target market is strategy departments at financial institutions, consulting firms, think tanks, and corporate research teams.
Sakana is positioning Marlin as the commercial distillation of several research threads: multi-model coordination for enhanced reasoning, automated scientific discovery pipelines, and what the company calls long-horizon reasoning. The system uses multiple models in coordination rather than a single base model, a design the lab has published on but has not disclosed in full for the production system.
Pricing uses four tiers: a free pay-per-use tier for occasional use, and Pro, Team, and Enterprise subscriptions for higher-volume teams. The product entered public availability on its launch date after a closed beta that ran from April 2026 onward with roughly 300 participants drawn from strategy, market analysis, risk assessment, and competitive intelligence roles.
Where this sits in the autonomous research market. Marlin enters a category that already includes Perplexity’s deep-research mode, OpenAI’s Deep Research, and Google’s Gemini Deep Research, all of which run multi-step web-research pipelines and produce structured reports. Sakana’s differentiator, per its own announcement, is multi-model coordination and a longer autonomous run window. Eight hours is a substantially longer claimed horizon than most competitors advertise; OpenAI’s Deep Research typically completes in under thirty minutes. Whether the extended window produces proportionally better analysis, or simply reflects a slower pipeline, is a question the launch materials do not address.
The announcement is Sakana’s own. No independent evaluation of Marlin’s output quality against competitor products or against human analyst work was included in the launch materials, and no external benchmark results were cited. The beta involved approximately 300 professionals, which is a reasonable sample for a beta, but Sakana has not published aggregate quality metrics from that period.
For teams currently evaluating autonomous research tooling, the practical question is cost per insight rather than hours per run. A 100-page report is only useful if the analysis density justifies the time investment. Strategy and consulting buyers who piloted the beta have a data point; everyone else should treat the claims as the company’s own until third-party comparisons appear.
Source: Sakana AI company announcement, published June 15, 2026, at sakana.ai/marlin-release/.
