
Sakana AI is envisioning a future where artificial intelligence evolves naturally, mirroring the adaptive and collective intelligence of living systems to create foundation models deeply aligned with human and societal needs. Our mission is to build a sovereign AI ecosystem in Japan, empowering communities and industries by offering democratized access to cutting-edge AI technologies tailored to local contexts.
We harness nature-inspired algorithms such as evolutionary models and continuous thought systems to pioneer a new paradigm beyond traditional AI architectures, advancing discovery and reasoning capabilities with unprecedented autonomy and cooperation. By bridging deep research with practical applications, we are shaping AI that learns, adapts, and innovates continuously within the complex environments it serves.
At Sakana AI, we are dedicated to cultivating a future where intelligent systems not only support but drive meaningful transformation across sectors in Japan and beyond. Through collaboration with industry leaders, government partners, and the global AI community, we forge pathways for AI to evolve responsibly, sustainably, and in harmony with human progress.
Our Review
When we first heard about Sakana AI, we'll admit — the name had us curious. Turns out, "sakana" means fish in Japanese, and there's something poetic about a company swimming against the current of mainstream AI development. Founded in Tokyo just last year by a trio of heavy-hitters who left big tech to chase bigger ideas, Sakana AI isn't trying to build the next ChatGPT clone.
Instead, they're doing something we find genuinely fascinating: teaching AI to evolve like nature does.
The Evolution Revolution
Here's where things get interesting. While everyone else is throwing more compute at transformer models, Sakana AI is asking a different question: what if we let AI models breed and evolve like living organisms? Their "Evolutionary Model Merge" technique literally combines existing AI models using evolutionary algorithms — think digital Darwinism.
We've seen the results, and honestly, it's pretty clever. Instead of training massive models from scratch (which costs millions), they're creating specialized AI by mixing and matching what already works. It's like getting a custom-bred dog instead of starting with wolves.
Japan's AI Sovereignty Play
What really caught our attention is how Sakana AI positioned itself as Japan's answer to AI independence. They've got backing from virtually every major Japanese corporation you can name — Mitsubishi UFJ, Sony, NTT, the whole gang. Even the Japanese government is giving them access to national data centers.
This isn't just about building cool tech; it's about ensuring Japan doesn't get left behind in the AI race. Smart positioning, and the $200 million Series A they just raised suggests investors are buying into the vision.
The Wild Science Projects
Beyond the evolutionary breeding stuff, Sakana AI is working on some genuinely sci-fi concepts. Their "AI Scientist" automatically conducts research and makes discoveries. They're building "Darwin Gödel Machines" that improve themselves autonomously. These aren't just incremental improvements — they're swinging for entirely new paradigms.
We appreciate that they're thinking decades ahead, not just the next funding round. Sure, some of these projects might not pan out, but that's the point of true R&D.
Who Should Pay Attention
If you're fascinated by the intersection of biology and AI, Sakana AI is definitely worth following. Their research papers are genuinely innovative, and they're tackling problems that most AI labs won't touch because they're too risky or long-term.
For job seekers, they're hiring aggressively and don't require Japanese language skills for technical roles. Though you will need to move to Tokyo — which, let's be honest, isn't exactly a hardship.
Nature-inspired foundation AI models
AI Scientist for automated research discovery
Evolutionary Model Merge for domain-specific AI model breeding
Continuous Thought Machines for adaptive reasoning
Darwin Gödel Machine for autonomous AI self-improvement
AB-MCTS cooperative multi-model inference algorithm






