Skip to main content

Sakana AI

Category: Foundation Models / LLMs

A world-class R&D lab developing 'nature-inspired' foundation models that are efficient, domain-specific, and created through automated evolutionary algorithms rather than massive-scale training. Sakana AI was founded in 2023. The company is led by David Ha. Based in Tokyo, Japan. Team size: 50-100. Total funding raised: ~$379M (¥30B Series A + ¥20B Series B). Latest round: Series B (¥20B/$135M, Nov 2025) at $2.65B valuation. Key investors include ["New Enterprise Associates (NEA)","Khosla Ventures","Lux Capital","NVIDIA","Google"].

Founded
2023
Headquarters
Tokyo, Japan
Team size
50-100
Total funding
~$379M (¥30B Series A + ¥20B Series B)

Value proposition

Delivers high-performance, culture-specific AI models with significantly lower compute and energy costs by 'breeding' existing models (Evolutionary Model Merge) instead of training from scratch.

Products and solutions

["EvoLLM-JP (Japanese Large Language Model created via evolutionary merging)","EvoSDXL-JP (High-speed Japanese Image Generation Model)","EvoVLM-JP (Vision-Language Model for Japanese context)","The AI Scientist (Fully automated scientific discovery system that generates hypotheses, writes code, and authors papers)","Llama-3-Evo-8B-Japanese (State-of-the-art Japanese LLM)"]

Unique value

Pioneered 'Evolutionary Model Merge,' a technique that uses genetic algorithms to automatically combine and optimize open-source models into superior, domain-specific models without heavy training.

Target customer

Japanese Enterprises (Finance, Manufacturing), Government Agencies (Sovereign AI), and Global Research Institutions.

Industries served

["Generative AI Infrastructure","Scientific Research & R&D","Financial Services (FinTech)","Government & Public Sector (Sovereign AI)"]

Technology advantage

Achieves State-of-the-Art (SOTA) performance with a fraction of the compute resources and data required by competitors (like OpenAI or Google), enabling rapid iteration and deployment on edge devices.

How they differentiate

Proprietary 'Evolutionary Model Merge' technique creating efficient, nature-inspired models; strong 'Sovereign AI' focus for Japan.

Main competitors

["Preferred Networks","Mistral AI","SoftBank (SB Intuitions)"]

Key partnerships

["Google (Strategic partnership for infrastructure/TPUs and R&D collaboration)","NVIDIA (Research collaboration and GPU infrastructure)","MUFG, SBI Group, Dai-ichi Life (Financial sector deployment)","KDDI & SoftBank (Telecommunications and distribution)","METI (Japanese Ministry of Economy, Trade and Industry - 'GENIAC' project support)"]

Notable customers

["METI (Japanese Gov)","KDDI","MUFG","SBI Group"]

Major milestones

["Released 'The AI Scientist' (Aug 2024)","Secured strategic investment from NVIDIA (Sept 2024) and Google (Jan 2025)"]

Growth metrics

Fastest Japanese startup to reach Unicorn status (~1 year); Valuation ~$1.5B.

Market positioning

Leader in efficient, domain-specific AI; Japan's national AI champion.

Geographic focus

Japan, Global (Research)

Patents and IP

Proprietary algorithms for 'Evolutionary Model Merge' and 'Automated Scientific Discovery' (The AI Scientist framework); specific patent filings are undisclosed but core IP lies in their automated model-breeding stack.

About David Ha

Former Research Scientist at Google Brain (Tokyo). Renowned for pioneering work in evolutionary computing and 'World Models'. Previously a Managing Director at Goldman Sachs (16+ years in finance).

Official website: