Sakana AI

Tokyo-based Sakana AI develops nature-inspired foundation models. Sakana AI's technology is based on collective intelligence and character-level training, aiming to build more adaptive and resource-efficient AI systems. The company adopts this technique by developing numerous smaller AI models that collaborate, much like a swarm, to deliver complex results. This is reflected in the name “Sakana,” which means “fish” in Japanese, reflecting the company's focus on collective intelligence in systems found in nature such as schools of fish and beehives.

In March 2024, the company released three AI models EvoLLM-JP, EvoVLM-JP, and EvoSDXL-JP, developed using a "model merging" technique, which combines existing AI models to form a new one. These comprise Japanese LLM, which is capable of math reasoning, and a Japanese vision language model (VLM). Two of these models are open-sourced and available on platforms like Hugging Face and GitHub.

In August 2024, Sakana AI launched 'The AI Scientist', a system for fully automatic scientific discovery to enable Large Language Models (LLMs) to perform independent research.

Key customers and partnerships

The company has formed an R&D partnership with Nippon Telegraph and Telephone Corporation (NTT), working on AI constellations—an architecture that connects numerous small AI models to work together as an intelligent collective. This collaboration aims to address the high computational demands of large AI models by distributing the workload across specialized, lightweight models.

In September 2024, Sakana AI partnered with NVIDIA to focus on research collaboration utilizing NVIDIA's GPU technologies, early access to NVIDIA-supported data centers in Japan, and developing Japan's AI community.

Key competitors: None comparable (Sakana AI is the only company in our coverage to focus on collective smaller models to deliver complex results.)