Sakana AI
Tokyo AI lab building nature-inspired foundation models
Overview
Tokyo AI lab founded in 2023 by ex-Google Brain's David Ha, Transformer co-author Llion Jones and Ren Ito, pursuing evolutionary approaches to foundation models. Now Japan's most valuable AI startup at $2.65B, shipping commercial products like the Marlin research agent.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 8-hour autonomous research run
- 60-100 page report + slides
- Add-on credits ~$0.61 each
- ~20 reports/month
- Priority queue
- Free 2nd month if signed up by Jul 31, 2026
- ~60 reports/month
- Multi-seat access
- Team collaboration
- Custom credit volume
- Dedicated support
- Contractual data-use terms
Use cases
What you can produce with Sakana AI
- A 60–100 page strategy research report with 60–80 cited sources and presentation slides (via Sakana Marlin)
- A fine-tuned Japanese-language LLM merged from open-source models without gradient training (via EvoLLM-JP recipe)
- An AI-generated scientific manuscript ready for workshop peer review (via AI Scientist v2 pipeline)
- A self-improving coding agent that iteratively rewrites its own code to hit higher SWE-bench scores (via Darwin Gödel Machine)
- A multi-model orchestration workflow for complex reasoning or coding tasks beating single-model baselines (via Sakana Fugu / TRINITY)
- Japanese-language conversational AI responses via the free Sakana Chat web interface
- Open-source AB-MCTS tree search implementation for custom LLM inference-time scaling (via TreeQuest on GitHub)
ASEAN Perspective
Sakana AI in Southeast Asia
Sakana AI is a Japan-headquartered company focused on Japanese institutional customers, with ASEAN coverage currently indirect. The company explicitly targets Japan's financial, defence, and manufacturing sectors with its Namazu LLMs and Marlin enterprise agent. A February 2026 partnership with Datadog — initially focused on Japan with plans for global expansion — includes data residency within Datadog's Tokyo data centre, making it relatively well-positioned for Japanese PDPA/APPI compliance. For ASEAN buyers in Singapore, Malaysia, or Indonesia, Sakana's products are largely inaccessible or untranslated today, though its open-source models (EvoLLM-JP, AI Scientist-v2) are freely downloadable and used by researchers across APAC. The $2.65 billion valuation and In-Q-Tel backing signal strategic ambitions beyond Japan, but no concrete ASEAN localisation roadmap has been announced as of mid-2026.
Sakana is one of the more technically credible AI labs outside the US and China -- its AI Scientist paper was the first fully AI-generated research to pass peer review, and the Darwin Godel Machine demonstrated a model that rewrote its own code and doubled its SWE-bench score. That research reputation is real, backed by a $2.65 billion valuation and money from MUFG, Nvidia and Lux Capital.
The commercial side is much younger. Sakana Marlin, its enterprise "virtual CSO" research agent, only launched in June 2026 off a roughly 300-person closed beta -- real-world reliability at scale is still unproven. Pricing runs enterprise-only (single runs from $66, Pro around $1,000/month), the product is Japanese-market-first, and there's no public API for third-party developers. A serious bet for organizations wanting Japan-local AI, not yet a drop-in general tool.
What people say
The AI research community rates Sakana highly, and for good reason: AI Scientist v2 became the first fully AI-authored paper to clear peer review at an ICLR workshop, and the Darwin Godel Machine's self-rewriting, self-improving code loop is the kind of result labs twice its size haven't shipped. None of that shows up on G2 or Capterra, because Sakana barely sells anything a typical SaaS buyer would review yet.
That's changing. Sakana Marlin, launched June 15, 2026, is the company's first real commercial product -- an autonomous research agent that runs for up to eight hours and returns a 60-100 page strategy report with cited sources and slides, built on Sakana's own AB-MCTS search algorithm. Pricing is steep and enterprise-shaped: pay-per-run starts around $66, a Pro subscription runs roughly ¥150,000 ($1,000) a month, Team tiers run higher, and Enterprise is quote-only. Tech coverage from TechCrunch, VentureBeat and MarkTechPost has been positive on the technical ambition but consistently flags that Marlin is fresh out of a roughly 300-person closed beta, with production reliability and support at scale still to be proven.
Sakana's pitch to Japanese finance, consulting and defense clients -- cultural alignment, no training on customer data by default, Tokyo-local infrastructure -- is a genuine differentiator regional buyers care about. But competing deep-research agents from OpenAI and Perplexity are more mature, cheaper per run, and globally supported. Sakana reads as an exciting, technically serious bet on AI sovereignty for Japan specifically, not yet a mature product for anyone evaluating on track record alone.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- Llion Jones, CTO and co-founder, was one of eight authors of the 2017 'Attention Is All You Need' paper that introduced the Transformer architecture underpinning all modern LLMs.
- Sakana means 'fish' in Japanese — the company logo and philosophy are inspired by how a school of fish collectively forms intelligent behaviour from simple individual rules with no central coordinator.
- Sakana AI's Darwin Gödel Machine improved its own SWE-bench score from 20% to 50% simply by rewriting its own code — no human intervention required.
- At its $2.65 billion November 2025 valuation, Sakana became Japan's most valuable private AI startup, surpassing rivals despite having fewer than 200 employees.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Sakana AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Sakana AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Sakana AI directly →
Spotted something out of date? Suggest an update →
Sakana AI in the news
Developer Tools
WWDC 2026 for Developers: An Expanded Foundation Models Framework and an Agentic Xcode 27
Developer Tools
Miasma: How a Hijacked CI Pipeline Shipped Malicious npm Packages With Valid Provenance
Developer Tools
GitLab 19.0 drops Redis for Valkey by default and brings Gitaly to Kubernetes
More in LLMs & Chat
