Sakana AI, a Japan-based AI startup, has released its multi-agent AI, Sakana Fugu, to the public. The system integrates multiple AI models to process tasks. This article highlights how it selects among the models and its plans for adding more in the future.
Sakana AI released its multi-agent AI system, Sakana Fugu, to the public on 22 June. Sakana Fugu selects the optimal model for each use case, using Claude for coding, GPT for reasoning and Gemini for writing, for example, then integrates the results to produce a final answer. The company says this can deliver higher performance than using a single model alone.
Sakana Fugu also determines the difficulty and type of each request and automatically routes it for processing. Another feature is that users can select or exclude the models to be used themselves. The company says benchmarks at launch showed that it outperformed Claude Mythos and Claude Fable 5 on certain measures. It is offered in three plans, Standard at $20, Pro at $100 and Max at $200, with pricing in dollars in anticipation of a global rollout.