Sakana AI
Sakana AI helps enterprises build specialized foundation models using evolutionary algorithms instead of massive compute.
Sakana AI develops foundation models using evolution-inspired techniques and collective intelligence to create domain-specific AI systems with minimal computational overhead. The company automates the entire AI development lifecycle through products like AI Scientist (automated research discovery) and Sakana Fugu (multi-agent model orchestration), serving enterprises in finance, defense, and research. Unlike traditional approaches requiring massive GPU clusters, Sakana's evolutionary model merge technique discovers optimal model combinations through natural selection, dramatically reducing energy consumption while achieving state-of-the-art performance.
Problem solved
Enterprise AI development currently requires prohibitive computational resources and energy consumption; Sakana enables creation of state-of-the-art foundation models through efficient evolutionary techniques without gradient-based training.
Target customer
Enterprise organizations in finance, defense, and research sectors seeking domain-specific AI models with reduced computational costs; Japanese financial institutions and government agencies.
Founders
D
David Ha
CEO
Hong Kong-born Canadian AI researcher; former Research Scientist at Google leading Brain Japan team; worked eight years as derivatives trader at Goldman Sachs; PhD from University of Tokyo, MS Mathematical Finance and BS Engineering Science from University of Toronto.
L
Llion Jones
CTO
Co-author of 'Attention Is All You Need' seminal paper at Google (2017); Computer Science and AI graduate from University of Birmingham with Master's in Advanced Computer Science.
R
Ren Ito
COO
Former Japanese ambassador; led European expansion of Mercari, a major online secondhand marketplace.
Funding history
Seed
$30M
January 2024
Led by Lux Capital
· Khosla Ventures, NTT Group, KDDI CVC, Sony Group, 500 Global, Miyako Capital, Basis Set Ventures, JAFCO, July Fund, Geodesic Capital, Learn Capital
Series A
$200M
September 2024
Led by NEA, Khosla Ventures
· Lux Capital, Translink Capital, 500 Global, NVIDIA, MUFG, SMBC, Mizuho, NEC, SBI Group, Dai-ichi Life Insurance, ITOCHU, KDDI, Fujitsu, Nomura Holdings, ANA Holdings, Tokyo Marine Group, Global Brain, JAFCO, Miyako Capital
Series B
$135M
November 2025
Led by Khosla Ventures
· Mitsubishi UFJ Financial Group, Factorial, Macquarie Capital, Fundomo, Mouro Capital, MPower Partners, NEA
Total raised:
$479M
Pricing
Not publicly available. Sakana Fugu accessible via APIs with OpenAI-format endpoint compatibility; product in beta testing phase.
Notable customers
MUFG, Daiwa Securities Group, MIT, OpenAI, Swiss AI Lab IDSIA
Integrations
OpenAI API format compatibility, HuggingFace (EvoLLM-JP, EvoVLM-JP release), GitHub, research partnerships with MIT and OpenAI
Tech stack
Open Graph
Google Analytics (Analytics)
Google Font API (Font scripts)
Varnish (Caching)
Google Workspace (Email)
Fastly (CDN)
GitHub Pages (PaaS)
Website
Competitors
OpenAI
OpenAI develops large monolithic models requiring massive compute; Sakana uses evolutionary techniques to efficiently adapt existing models for specific domains with less energy overhead.
Anthropic
Anthropic builds custom foundation models through traditional training; Sakana merges and evolves existing models through evolutionary algorithms without gradient-based training.
Together AI
Together AI provides inference infrastructure for existing models; Sakana automatically discovers optimal model combinations and develops domain-specific foundation models through evolution.
Why this matters: Sakana AI achieved a landmark milestone in March 2025 when its AI Scientist system produced the first peer-reviewed paper accepted at a premier machine learning conference written entirely by AI, validating its automated research discovery approach. The company combines world-class talent (including the co-author of 'Attention Is All You Need') with a novel evolutionary approach to AI development that directly addresses the sustainability and cost challenges plaguing the industry.
Best for: Enterprise organizations in finance, defense, and research needing specialized AI models that require lower computational resources and faster development cycles than traditional foundation model training.
Use cases
Domain-Specific Financial AI
Financial institutions like MUFG and Daiwa Securities use Sakana to develop custom foundation models optimized for Japanese finance operations, regulatory environments, and tacit business knowledge without training massive models from scratch.
Automated Scientific Discovery
Researchers use AI Scientist to automate the entire research lifecycle—from generating novel ideas and writing code to executing experiments and writing manuscripts—enabling faster scientific advancement without manual research cycles.
Multi-Agent Model Orchestration
Enterprises deploy Sakana Fugu to coordinate multiple frontier foundation models for complex tasks spanning coding, mathematics, and reasoning, achieving state-of-the-art performance across domains without maintaining separate model infrastructure.
Alternatives
OpenAI API
General-purpose closed-source models requiring large compute; Sakana offers domain-specific models using efficient evolutionary techniques with greater customization for specialized applications.
Hugging Face
Model hub providing pre-trained models; Sakana actively evolves and merges models for specific domains using proprietary evolutionary algorithms rather than passive model distribution.
LLaMA/Meta
Open-source base models requiring fine-tuning; Sakana's approach optimizes existing models through evolutionary techniques without expensive retraining, reducing compute requirements.
FAQ
What does Sakana AI do? +
Sakana AI develops specialized foundation models using evolution-inspired techniques and collective intelligence. The company offers AI Scientist (automated research discovery), Sakana Fugu (multi-agent model orchestration), and domain-specific models for finance and other sectors. Unlike traditional approaches requiring massive compute, Sakana's evolutionary model merge discovers optimal model combinations through natural selection.
How much does Sakana AI cost? +
Pricing is not publicly available. Sakana Fugu is accessible via APIs with OpenAI-format endpoint compatibility. Contact Sakana directly for enterprise pricing and custom deployment options.
What are alternatives to Sakana AI? +
Alternatives include OpenAI API (general-purpose closed-source models), Hugging Face (model distribution hub), and Meta LLaMA (open-source base models). Sakana differs by focusing on efficient evolution of domain-specific models rather than training monolithic foundations or providing generic model distribution.
Who uses Sakana AI? +
Target customers include enterprises in finance (MUFG, Daiwa Securities), defense/government sectors, and research institutions. Japanese financial institutions and government agencies are primary early adopters seeking domain-specific models tailored to local business practices and regulatory requirements.
How does Sakana AI compare to OpenAI? +
OpenAI develops large closed-source models requiring enormous computational resources. Sakana uses evolutionary techniques to efficiently adapt and merge existing models for specific domains with significantly lower energy overhead. Sakana targets enterprise customization, while OpenAI emphasizes general-purpose APIs.
Tags
foundation models
evolutionary algorithms
AI research automation
domain-specific models
model optimization
Japanese AI
enterprise AI