Sakana AI has updated its Fugu Ultra model router to version 1.1, claiming gains of up to 7.9 points over the previous release. The Decoder notes that independent verification is not yet available.
A model router chooses which AI model should answer a given request. If it works well, users can get stronger results or lower costs by sending easy, hard, coding, or reasoning tasks to different models instead of relying on one default system.
The update also adds a Claude Code-compatible endpoint, which could make it easier for developers to test the router inside existing coding workflows. Sakana says the service remains unavailable in the European Union.
The useful part of the announcement is the workflow direction: routing is becoming a product layer above individual models. The limitation is equally important. Claims about beating another model or router need outside testing on real tasks before teams rely on them for cost or quality decisions.