Sakana dropped Fugu Ultra and the announcement went viral claiming it matches Fable and Mythos. Fugu isn't a new model. It's one API that orchestrates and routes tasks across frontier models like Opus, GPT, and Gemini, kind of like Claude Code spinning up sub-agents, just automatic.
So I ran Fugu Ultra against Claude Opus 4.8 across 38 tasks to see if the orchestration actually wins. In this video you'll see where it landed on quality, speed, and cost, and why I'm not switching off my Claude Code and Codex subscriptions yet.