Sakana AI Ships Fugu-Ultra v1.1 With a Claude Code Interface for Its Multi-Agent Orchestrator
A month after launching Sakana Fugu, Sakana AI shipped Fugu-Ultra v1.1 on July 24, 2026, pairing meaningful benchmark gains with a new Claude Code-compatible interface that puts Fugu’s multi-agent orchestration inside a terminal workflow developers already use.
Details
- What improved: Sakana reports gains of up to 7.9 points over Fugu-Ultra v1.0, with the largest jumps on ProgramBench and Terminal Bench 2.1 — both agentic coding benchmarks
- Competitive claim: Sakana states that v1.1 now outperforms rival model Fable 5 on complex coding and reasoning tasks (a claim from Sakana’s own post, not independently benchmarked here)
- No price change: the upgrade ships at the same pricing as v1.0, with no separate v1.1 tier
- Claude Code integration: Fugu now exposes Claude Code-compatible endpoints, so developers can point Claude Code at Fugu’s orchestrated backend — a coordinated team of underlying models — instead of a single model, for writing, debugging, and running code from the terminal
- Positioning: rather than compete as a single model, Sakana is leaning into orchestration as the differentiator — the pitch is that a coordinated committee of models plugged into an existing tool (Claude Code) beats a single frontier model on hard agentic tasks
- Quote: “The frontier keeps moving, and Fugu keeps getting better.”
What happened next
- Existing Fugu API customers received the v1.1 upgrade automatically at their existing pricing
- Developers can set up the Claude Code interface via Sakana’s console at console.sakana.ai/get-started
- Product page: sakana.ai/fugu; full announcement: sakana.ai/fugu-1-1-claude-code-interface