Japanese AI startup Sakana AI has launched Fugu (and Fugu Ultra), a multi-agent system disguised as a single model, accessible via an OpenAI-compatible API. • Key Features: • Instead of training a massive model from scratch, it smartly orchestrates multiple cutting-edge existing models (including recursive calls to itself or others like GPT, Claude, Gemini, etc.). • Automatic task routing: selects the right agent, verifies results, synthesizes outputs. • Fugu: Designed for quick, everyday tasks with low latency. • Fugu Ultra: Tailored for complex reasoning, coding, and scientific tasks, with performance reportedly rivaling or even surpassing high-end models like Claude (excelling in benchmarks like SWE-Bench, GPQA, etc.). • Highlights: Bypasses U.S. export controls (no reliance on a single supplier), making it ideal for enterprises and developers seeking resilience. Seen as an innovative direction for "orchestration-as-a-product." • Current feedback: Fast and flexible, but benchmark authenticity is under discussion—best suited for real-world testing. https://(x.com)/SakanaAILabs/status/2068861630327443966/photo/1