Sakana AI's Fugu: Coding & Reasoning Benchmarks Leader

Japan's Sakana AI has unveiled a new multi-agent orchestration platform called Fugu. The company claims Fugu surpasses Anthropic’s Claude Fable 5 on certain coding and reasoning benchmarks. Fugu comes in two versions: a standard model for everyday tasks like coding and chat, and Fugu Ultra for more complex work. On LiveCodeBench, Fugu Ultra scored 93.2 and the standard Fugu model scored 92.9. Anthropic’s Claude Fable 5 recorded 89.8, while OpenAI’s GPT 5.5 scored 85.3. Fugu and Fugu Ultra also achieved the highest score of 95.5 on GPQA-Diamond, a benchmark for scientific reasoning. Early users report Fugu Ultra outperformed GPT-5.5 in code review, finding bugs other models missed. This new platform highlights the ongoing competition in AI development.