ChainThink report: On June 25, Sakana AI, a Japanese AI startup, launched its multi-agent orchestration system Fugu Ultra on OpenRouter.

The system dynamically orchestrates GPT-5.5, Gemini 3.1 Pro, and Claude Opus 4.8 via a single interface, achieving benchmark scores surpassing Anthropic’s flagship model Claude Fable 5 and Mythos Preview in academic and programming evaluations.

Fugu Ultra is a language model specifically trained in coordination mechanisms, drawing from the ICLR 2026 papers Trinity and Conductor. It enables fine-grained, dynamic role allocation based on context: Gemini 3.1 Pro acts as the aggregator in common-sense QA tasks, GPT-5.5 assumes the aggregator role for bias correction in mathematical computations, and in multi-turn coding scenarios, GPT-5.5 handles code generation while Claude Opus 4.8 performs audit and debugging alternately.

Currently live on OpenRouter, the system offers 1 million token context length and up to 128,000 token maximum output, supporting tool calling and multimodal input.

Pricing is $5 per million input tokens, $30 per million output tokens, and $0.5 for cache reads;

the underlying model consumption generated by each API call is included in the total cost of orchestration and output; users can also control reasoning depth and maximum token limits via the reasoning parameter.