Sakana's multi-agent orchestrator makes its way into Claude Code

Sakana's Fugu orchestrator partners with Claude Code. The multi-agent approach applies to debugging and writing code directly in the terminal.

Already presented as a multi-agent orchestrator delivered as a single model, Sakana Fugu is receiving two updates. Its high-end engine, Fugu-Ultra, moves to version 1.1 with an agent pool updated to newer frontier models, at no additional cost. Most notably, the publisher is launching a Claude Code-compatible interface.

This compatibility makes it possible to call Fugu's multi-agent intelligence from the terminal, inside Claude Code. Rather than entrusting code writing, debugging, and execution to a single model, the request is sent to Fugu, which distributes the work among several coordinated state-of-the-art models before returning a synthesized response. Since Fugu has been accessible via platforms like OpenRouter and Vercel since its launch, this addition extends the same distribution logic, this time plugging it into a widely used development environment.

The Fugu-Ultra update focuses on reasoning, code, and agentic tasks, with the publisher claiming gains on the benchmarks it tracks, with no price change compared to the previous version.