Fugu and Fugu-Ultra are now selectable in OpenCode

Sakana AI's Fugu can now be chosen as the model behind OpenCode, an open-source coding agent that works in the terminal. Fugu is a paid service that accepts an ordinary model request and passes the work to a coordinating system that can consult several other AI models before returning one answer. Sakana's July 2, 2026 post announces the availability and says the team used OpenCode as its tool for verifying models while it was building Fugu's multi-agent orchestration. The image attached to the post shows only the two logos, so the post itself names no specific models and gives no setup steps.

The model choices come from the registry OpenCode reads. OpenCode's provider documentation says it uses the AI SDK and Models.dev to support its providers, and the Models.dev catalog, checked on October 31, 2026, lists a Sakana AI provider that reads a SAKANA_API_KEY, calls Sakana's OpenAI-compatible endpoint, and offers Fugu and Fugu Ultra. A developer with a Sakana key can therefore select either one inside OpenCode instead of writing code against the API. That catalog check is four months later than the post, so it confirms the current listing rather than exactly what appeared on launch day.

The two entries differ in how much coordination they buy. Sakana's product page describes Fugu as the balanced default for everyday coding and Fugu Ultra as coordinating more expert agents for complex, multistep work at the cost of response time. The post makes no performance claim for Fugu inside OpenCode, and the remark about verifying models describes Sakana's internal practice, not a published evaluation. Our explanation of the Fugu launch covers how the coordinating model works.