Intel / article

Sakana's Fugu-Ultra v1.1 can now drive Claude Code and Codex

Source: console.sakana.ai
Author: Sakana AI
Date: Published Jul 26, 2026

Context

Sakana AI made its Fugu service usable as the model inside two existing terminal coding agents, Anthropic's Claude Code and OpenAI's Codex CLI. Fugu takes one model request and hands the work to a coordinated pool of models from several providers. Connecting it to these agents means a developer keeps the 's file editing, command running and subagents while the reasoning comes from Fugu. Sakana's July 26, 2026 post announced the Claude Code-compatible interface together with Fugu-Ultra v1.1, and its setup guide explains how the connection works.

The connection relies on Sakana's API accepting each agent's native request format. Claude Code can be redirected with environment variables: ANTHROPIC_BASE_URL points it at api.sakana.ai, a Sakana key goes in ANTHROPIC_AUTH_, and further variables map Claude Code's model tiers onto Fugu models, with the Opus tier going to Fugu Max, the Sonnet and Haiku tiers and subagents to Fugu, and the Fable tier to Fugu Ultra. For Codex, a provider entry in its config.toml sets Sakana as the endpoint using OpenAI's Responses format and raises the stream idle timeout to two hours, because Codex otherwise drops a turn after roughly five idle minutes, and Fugu's slower orchestrated answers can exceed that. A one-line installer for Ubuntu and macOS sets both up and adds claude-fugu and codex-fugu launchers.

Sakana says the Claude Code side has limits because Claude Code is closed source. Some interface elements still reflect Anthropic's assumptions: the effort slider's six stops collapse onto Fugu's two real reasoning levels, and older clients and manual setups show duplicate rows in the model picker. Sakana describes these mismatches as cosmetic and says responses, , and subagents behave as configured. Separately, the claude-fugu launcher does not update itself, so new Fugu models require a manual update. The guide read here is a later version than the July announcement: it now defaults to Fugu Max and lists Fugu Ultra v1.1 only as an optional pinned choice. Sakana's performance claims for v1.1 appear in its release article and are not evaluated here.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.

Read the source console.sakana.ai