Japanese artificial intelligence startup Sakana AI has unveiled Fugu, a new family of AI systems that combines multiple foundation models through an orchestration framework designed to improve performance across coding, scientific reasoning, mathematics, and complex analytical tasks. The company describes Fugu as a multi agent platform capable of coordinating several AI models to solve problems more effectively than relying on a single model.

According to Sakana AI, Fugu achieved strong results across several industry benchmarks used to evaluate software engineering and reasoning capabilities. The company claims that its advanced version, Fugu Ultra, delivered competitive or superior performance compared with leading frontier AI models on selected coding and scientific reasoning tests.

The launch comes amid growing competition among AI developers including Anthropic, OpenAI, Google, and emerging AI startups. Anthropic recently introduced Claude Fable 5, which it described as its most capable publicly available model with strong performance in software engineering, scientific research, and knowledge work.

Sakana AI's approach differs from traditional large language models by focusing on orchestration, where multiple AI models collaborate under a central coordination system. The company believes this collective intelligence framework can produce better results while reducing costs compared with building ever larger standalone models.

The introduction of Fugu highlights the rapid evolution of the artificial intelligence sector, where developers are increasingly exploring agent based and multi model architectures to improve performance in specialized tasks such as software development, scientific research, and advanced reasoning. While Sakana AI's benchmark claims have attracted attention, broader independent validation of performance comparisons with competing models remains an important factor for industry adoption.