Asia's New Powerhouses: Japan and China Rush to Replace America's AI Dominance

Japanese artificial intelligence (AI) startup Sakana AI launched a new generation of AI systems called Fugu and Fugu Ultra on Tuesday (23/6). Unlike conventional AI models that rely on a single large model, this latest technology works by coordinating multiple AI models simultaneously to complete various tasks. The Tokyo-based startup claims Fugu Ultra matches the performance of Anthropic’s Fable 5 and Mythos 5 in key benchmarks for engineering, science, and reasoning. On certain tasks, Fugu Ultra is said to surpass Fable 5. Anthropic’s Fable 5 and Mythos 5 were recently withdrawn after the US government requested the company block access on national security grounds. The Fugu version is intended for programming, conversation, and daily tasks, while Fugu Ultra is designed for complex work such as AI research, reproducing scientific papers, cybersecurity analysis, and patent investigation. According to benchmark graphs shared by Sakana, Fugu outperforms Claude Fable 5 on LiveCodeBench, an open-source benchmark testing programming abilities, with Fugu Ultra scoring 93.2 and Fable 5 scoring 89.8. Fugu also surpassed Claude Mythos on the GPQA-D (Diamond) test, which consists of 198 postgraduate-level multiple-choice questions in biology, physics, and chemistry. Sakana claims its models also outperform Google’s Gemini 3.1 Pro, OpenAI’s GPT-5.5, and Anthropic’s Opus 4.8 in tasks including automated research, mechanical design, Japanese handwriting analysis, one-shot chess, Rubik’s Cube solving, and financial time series prediction. Sakana argues that instead of relying on a single model, Fugu combines the strengths of various AI models. The company was founded in Tokyo in 2023 by Llion Jones, a co-author of Google’s seminal 2017 paper ‘Attention Is All You Need’, and David Ha, former head of research at Stability AI. The launch sparked discussion in the AI community about shifting focus from building ever-larger models to coordinating existing systems. Box CEO Aaron Levie noted the innovation effectively routes tasks to the best-performing model. However, Wharton professor Ethan Mollick tested the system and found it ‘very slow’, noting a coding test took 30 minutes and did not match Fable in real-world use. Meanwhile, Chinese cybersecurity firm 360 Security Technology announced it has developed its own AI tools, introducing ‘Yitian Tulong’ at the ISC.AI 2026 conference in Beijing. Founder Zhou Hongyi described one tool, Tulongfeng, as China’s version of Mythos, designed to automatically discover software vulnerabilities. He argued that such powerful cyber weapons must not be held exclusively by other nations, citing the risk of one-way transparency if US entities can scan critical Chinese software without equivalent Chinese capability. The company claims Tulongfeng has already discovered 3,432 software vulnerabilities.