In the bustling tech hubs of Tokyo, a startup named Sakana AI is redefining artificial intelligence by drawing inspiration from the natural world, eschewing the brute-force scaling that dominates Silicon Valley. Founded in 2023 by former Google researchers David Ha and Llion Jones, the company has rapidly ascended to unicorn status, valued at over $1 billion, making it Japan’s fastest-growing AI venture. Their approach mimics schools of fish—simple entities forming complex, coherent systems through collective behavior—aiming to create efficient, adaptive AI without the massive energy demands of models like ChatGPT.

Sakana’s core innovation lies in evolutionary algorithms that “breed” AI models, allowing them to merge, compete, and evolve much like organisms in nature. This method, detailed in their recent work on “Competition and Attraction Improve Model Fusion” presented at GECCO’25, enables the automatic combination of existing open-source models to birth new ones with enhanced capabilities. As Sakana AI’s official site explains, this nature-inspired intelligence focuses on collective dynamics, where individual components interact to produce emergent intelligence, potentially slashing the computational costs that plague traditional AI training.

A Leap Beyond Conventional Scaling: How Evolutionary Merging is Reshaping AI Development

Recent advancements, such as the Evolutionary Model Merge technique released in 2024, have garnered attention for automating the creation of foundation models. By selecting and reordering layers from “parent” models or mixing their parameters, Sakana produces offspring models tailored to specific tasks, as highlighted in a SiliconANGLE report. This evolutionary process not only boosts efficiency but also leverages the “vast collective intelligence” of open models, reducing the need for expensive data centers—a critical edge in resource-scarce Japan.

The company’s “AI Scientist” project takes this further, using foundation models to automate scientific discovery. In a 2024 paper, Sakana described harnessing LLMs to generate new objective functions and merge knowledge autonomously, surprising even its creators with the creative output. According to Cogni Down Under on Medium, this positions Sakana at the forefront of open-ended research, where AI agents could independently brainstorm, code, and iterate on ideas, minimizing human oversight.

Brain-Like Timing and Continuous Thought: Mimicking Neural Dynamics for Smarter AI

Delving deeper, Sakana’s Continuous Thought Machine (CTM) introduces neuron-level timing and synchronization, emulating how the brain processes time for step-by-step reasoning. A May 2025 post on The Decoder notes that this biologically inspired model enhances interpretability and efficiency, blurring lines between human and machine cognition. Posts on X from users like @SakanaAILabs emphasize CTM’s potential for flexible intelligence, with one 2025 update describing it as a ground-up build using neural dynamics to tackle complex tasks.

This innovation extends to practical applications, such as the Model Merging of Natural Niches (M2N2), which augments AI without retraining costs. As reported in a recent Heartspace Industry News article, M2N2 enables developers to evolve models adaptively, drawing acclaim for its cost-effectiveness amid soaring AI energy demands.

Global Ambitions from Tokyo: Blending Cultures and Overcoming Challenges

Sakana’s Tokyo base, as Ha told Yahoo Finance in a September 2025 interview, reflects a strategic choice to foster unique AI in a democratic Asian hub, blending U.S. and Chinese technologies. Nikkei Asia’s coverage last week echoed this, with CEO insights on devising “unique Japanese AI” through hybrid systems. Despite Japan’s infrastructure hurdles, Sakana’s $30 million seed round in 2024 and Series A funding have fueled growth, including open-house events attracting hundreds, per their site.

Industry sentiment on X buzzes with excitement; posts from influencers like @hardmaru celebrate “AI creating AI,” while others highlight evolutionary “mating” of models as a game-changer. Yet challenges remain: ensuring ethical evolution and avoiding unintended biases in self-improving systems.

The Future of Collective Intelligence: Sakana’s Role in AI’s Next Paradigm

Looking ahead, Sakana aims to merge large and small systems into “collective intelligence,” as outlined in a Japan Times piece just hours ago. This could disrupt the AI arms race, offering sustainable alternatives to power-hungry giants. By 2025, with projects like the AI Scientist pushing boundaries, Sakana isn’t just innovating—it’s evolving the very essence of intelligence, proving that inspiration from nature might hold the key to AI’s sustainable future. As one X post from @dl_weekly noted, this evolutionary technique could end the era of costly fine-tuning, heralding a new wave of accessible, adaptive AI.