Sakana AI’s Fugu Orchestrates AI Models to Dodge Bans and Boost Performance

Sakana AI launched Fugu on June 23, 2026, just ten days after the US government forced Anthropic to pull its frontier models offline. Fugu is not a single AI model but an orchestration system managing a pool of models through one API. It aims to deliver cutting-edge AI results without relying on any single provider.
Fugu comes in two flavors: a faster version for coding and chat, and Fugu Ultra for heavier tasks like patent research and security testing. Both perform at or above the level of Anthropic’s banned Fable 5 and Mythos preview in several benchmarks. Fugu Ultra, model ID “fugu-ultra-20260615,” shines on complex tests, scoring 73.7 on SWE-Bench Pro and topping 10 out of 11 benchmark categories.
Unlike traditional monolithic models, Fugu Ultra isn’t bigger. It’s smarter. It orchestrates when to call each sub-agent, how they communicate, and how to merge their outputs. This training comes from Sakana’s peer-reviewed ICLR 2026 papers Trinity and Conductor. This design lets Fugu handle long, complex tasks—like playing four blindfold chess games simultaneously or designing a mechanical iris in CAD.
Fugu’s architecture routes each request across a swappable model pool, making it resilient to outages or bans of individual models. This “un-bannable” goal remains unproven but is central to its design. Its closed-source system partly relies on undisclosed closed APIs, and the composition of its model pool remains a secret. Early users report Fugu Ultra finds over twenty issues in code review where other tools spot only three.
Cost remains steep, with prices reaching around $10 per message for demanding tasks. Fugu isn’t launching in the EU, limiting its initial reach. Still, its ability to stand shoulder to shoulder with banned frontier models without the same export control risks makes it a unique player. As one director put it, “Viktor is now an integral team member, and after weeks of use we still feel we haven’t uncovered the full potential.”
Based on
- Sakana’s Fugu takes aim at the frontier — therundown.ai
- Sakana Fugu: On Par With Mythos! 🌷👀 — intheworldofai.com
- Sakana AI Launches Sakana Fugu: An Orchestration Model That Routes Tasks Across a Swappable Pool of Frontier LLMs – mGrowTech — mgrowtech.com
- Sakana Fugu Challenges Mythos — aiplanetx.com
- Sakana Fugu Ultra promises to deliver “the very best frontier-level performance” by… using frontier AI models. | The Verge — theverge.com



