Sakana Marlin is the new B2B product from Sakana AI, an agent that runs for eight hours before delivering a full strategy report. We threw a real SEO brief at it and read the hundred pages it produced. Verdict after the run.
Key Takeaways
- Marlin runs uninterrupted for eight hours through its AB-MCTS engine and outputs a structured report plus a slide deck ready to present
- Fed a real SEO brief, it delivered a dense competitive audit, a keyword cluster matrix and an editorial plan usable straight out of the tool
- Billing is credit-based, starting at ¥98 per credit with a hundred credits per run, putting one full study around ¥9,800
Have an AI Sum Up This Article
ChatGPTWhat Sakana Marlin actually does
Sakana AI positions Marlin as a Virtual CSO, a virtual chief strategy officer built for business teams. It is a B2B agent that takes a brief, runs autonomously for up to eight hours and delivers a main report, appendices, a full citations list and a slide deck at the end.
The core engine is called AB-MCTS, short for Adaptive Branching Monte Carlo Tree Search. It is an autonomous hypothesis-testing engine that explores several reasoning branches in parallel, drops the weaker ones and doubles down on the branches that surface solid evidence. In practice, Marlin does not write a linear summary. It builds a tree of strategic options.
The gap with Perplexity Deep Research or the ChatGPT Deep Research feature comes down to two things. First, run duration, eight hours versus a few minutes to an hour. Second, the output format, a structured menu of strategic options rather than a single narrative summary. These are different products with different jobs. We ran a similar hands-on elsewhere, our test of ChatGPT Plus at 20 dollars.
The target buyer is stated clearly, companies, organizations and sole proprietors running a real P&L. That is the same territory as the Meta Hatch agent priced at 200 dollars a month to do an employee’s work, only with a different angle, more research than production. Marlin does not handle operational tasks. It delivers structured thinking.
One last critical point, every claim in the final report is sourced. Marlin cites primary sources with live links, gathered into a dedicated bibliography at the end of the document. For a deliverable meant to land in front of a board or a client, that is a non-negotiable guarantee.
The walkthrough: handing it an SEO strategy
We built a brief tailored to SEO. A French B2B site in the AI and entrepreneurship space, five named competitors with their URLs, a goal of top three rankings on a shortlist of commercial keywords within six months, an existing editorial line and a known monthly production budget.
The brief fit on two pages. We specified the exact deliverables we wanted, a competitor-by-competitor SERP snapshot, a keyword matrix prioritized by intent and difficulty, a twelve-month content plan organized in clusters, a backlink gap analysis and a risk scenario.
Kicking off a run costs a hundred credits. There is no visible timer, you close the tab and come back the next day. Sakana Marlin sends interim pings on the branches it is exploring, which offers reassurance without forcing you to babysit the agent.
During the eight hours, the Sakana AI agent does what a senior consultant with three juniors would do. It scrapes the visible SERPs, cross-references with visible ad data, reverse-engineers competitor content architecture, identifies topic clusters that carry weight for them and not for us, then builds a game plan. It does not run a technical crawl on the target site, that is out of scope.
On the broader limits of autonomous agents, the Tencent study that measures a 14% completion rate on real-world tasks tempers expectations. Marlin is no exception, but its output format is deliberately built to be auditable, which makes the gap between promise and execution far more visible than with a browser-piloting agent.
Also on Horizon:
- Sakana AI: We Tested the Japanese AI Everyone Talks About
- SpaceX AI Phone Prototype Leaks, Musk Denies It All
- Meta Compute: Zuckerberg Rents Out Spare AI GPUs
The verdict after reading a hundred pages
First finding, the competitive analysis section is solid. For every competitor listed in the brief, Marlin produces a full editorial profile, priority topics, dominant formats, publication cadence measured across the last six months, standout backlinks and an estimated ratio of brand queries versus commercial queries.
The keyword matrix is where the raw value sits. Every term is ranked by intent, estimated search volume, assumed difficulty and editorial fit with the current line of the site. Keywords are grouped in thematic clusters, so the reader gets an actionable editorial plan from page one, not a flat list to process manually.
The content plan proposes a twelve-month cadence built around the clusters, one pillar article and a series of satellite pieces per cluster. Marlin also proposes a production sequence that factors in indexation waves. This is what a seasoned SEO consultant would formalize, only produced faster.
The weak spots are real. Marlin does not perform a technical on-page audit, it does not crawl site architecture, does not measure Core Web Vitals and does not plug into Search Console. It also does not propose a calendar anchored on real dates, only a theoretical cadence. On link-building recommendations, its output stays generic and lacks a shortlist of qualified opportunities with actual contact leads.
The slide deck is usable as is for an internal readout. Thirty to forty slides, clear structure, one slide per cluster, one slide per risk, a full roadmap slide. These are working slides, not sales-grade client slides, but that already saves a consultant a serious block of synthesis time.
Who Sakana Marlin actually fits
Sakana AI publishes pricing in yen. The Pay as you go tier costs ¥98 per credit, one study consumes a hundred credits, so a single study lands around ¥9,800. The Pro plan is ¥150,000 per month for two thousand credits, with extra credits billed at ¥90. The Team plan is ¥400,000 per month for six thousand credits, with additional credits at ¥85. Enterprise is custom-quoted with dedicated support.
For a freelance SEO consultant who sells a premium competitive analysis to a client, ¥9,800 per study, which is the equivalent of one day of work delivered overnight, changes the math. Margin per engagement holds while billable hours drop, and the deliverable stays presentable in a client meeting. That is the profile with the strongest immediate ROI.
For an in-house growth or marketing team briefing several topics a month, the Pro plan becomes profitable as soon as the pace exceeds two heavy briefs per month. The ¥90 overage credits handle spikes without triggering a new contract negotiation.
For an agency, the Team plan standardizes the upstream phase of production. Where three juniors would spend a week scoping an SEO strategy for a client, Marlin produces an auditable deliverable that a senior can challenge and refine in a day. The gain is real on time to delivery, not on unit margin, which stays captive to seniority cost.
The real debate is human replacement. Marlin replaces one week of a junior consultant’s work, not the judgement of a senior making calls. It thinks wide, structures fast and sources cleanly, but it does not judge. Final prioritization, the decision to cut a cluster or pivot the angle, still sits with the human who knows the business.
Bottom line, Sakana Marlin is an excellent generator of auditable strategic drafts. It is neither a senior consultant replacement nor an SEO production tool, but a scoping accelerator that puts a serious competitive analysis within reach of an overnight run and a junior-day budget.
Follow the story on Horizon.
