A Japanese AI company, Sakana, has introduced an AI system called “Fugu,” and multiple reports say it performs better than Anthropic’s Claude 5 on some benchmark tests. The coverage describes Fugu as a new system developed by Sakana and notes that comparisons are based on specific evaluations rather than an overall, universal measure of capability. While the reporting highlights the benchmark results and the head-to-head framing against Claude 5, the details available in the provided excerpts do not specify which benchmarks were used, the size of the margin, or whether the results reflect single runs or broader testing methodology. The accounts also do not indicate whether the benchmark set is limited to a particular type of task or model setting. Overall, the reporting presents the launch and the reported performance gains as a measured outcome on certain tests, while leaving open broader questions about generalization across different tasks and conditions.