Summary
Sakana AI, a Japanese company, has launched Fugu Ultra — a multi-agent orchestration system that routes tasks through multiple frontier models (Claude Opus, GPT-5.5, Gemini) via a single API. The company claims Fugu Ultra matches or outperforms models like Anthropic’s Fable and Mythos on select benchmarks by dynamically delegating subtasks to specialized sub-agents. Creator Nate Herk ran 38 real-world coding and productivity tests comparing Fugu Ultra head-to-head against Claude Opus 4.8 to pressure-test those claims.
The results tell a sobering story: 36 of the 38 tasks ended in a tie, with Fugu Ultra offering no meaningful quality advantage. More pointedly, Fugu was 4.5 times slower and five times more expensive than Claude Opus 4.8 alone. At the $200/month subscription tier, Herk found the weekly usage limits filled up quickly — significantly faster than a comparable Claude Code subscription.
The video also unpacks the architectural difference between Fugu and OpenRouter’s Fusion API, and contextualizes Fugu’s orchestration approach against what Claude Code already does natively with its own sub-agent system. The conclusion: for developers already using Claude Code, Fugu’s cross-model orchestration premium is hard to justify. For those wanting a single API that spans competing frontier models without managing routing themselves, it offers genuine convenience — at a cost.
📺 Source: Nate Herk | AI Automation · Published June 23, 2026
🏷️ Format: Review
