Sakana Fugu: AI Routes Tasks to Best LLMs for Efficiency

Sakana AI has launched Sakana Fugu, a new AI system that acts like a smart manager. It routes each task to the best AI model from a pool and then combines the results. Here's the thing: most AI products use one model for everything. Fugu, however, is an "orchestration model" that coordinates many parts. It sits behind a single API endpoint, making it easy for developers to integrate. What's interesting is that Fugu itself is a language model. It's trained to choose which model to use, hand off the task, check the answer, and combine results. It learns the best way to delegate without fixed rules. The "swappable pool" means Fugu can use new, better models as they appear without needing a full rebuild. Sakana offers two versions: Fugu, which balances performance with low latency for tasks like coding and chatbots, and Fugu Ultra, tuned for the highest quality on difficult, multi-step problems. For example, Fugu Ultra scores 73.7 on real coding fixes, compared to 69.2 for Opus 4.8. The bottom line: this new approach could make AI systems more flexible and efficient by leveraging the strengths of multiple models.