Sakana AI released Fugu-Ultra v1.1 on July 23, posting the update in a thread on X. The company says the new version is more capable at coding, agentic tasks, and advanced reasoning than Fugu-Ultra v1.0, and it ships at the same price as its predecessor.
Fugu is built around orchestration rather than a single trained model. Sakana AI describes it as pulling together multiple underlying models to work through complex, multi-step tasks, arguing this approach delivers frontier-level results without tying a product to one vendor’s roadmap. That is an architectural claim, not just a features claim: orchestrating several models is being pitched as superior to depending on any single one.
Sakana AI has not published independent benchmarks for these gains; every capability claim in the thread is the company’s own. The pitch also arrives the same week routing and multi-model products shipped from other vendors, which makes the underlying architecture, not the version number, the story worth tracking.
Announced by Sakana AI, on X, July 23, 2026.
