Xinyu Yang, a researcher at Moonshot AI (Kimi), expressed shock at the thousands of GPUs available to researchers at OpenAI and other US frontier labs, noting this compute scarcity makes frontier innovation in China incredibly difficult. The post also lists current model rankings, including Fable5, GPT 5.6 sol, Kimi K3, Grok 4.5, and GLM 5.2, and suggests Zhipu's GLM 5.2 has significant potential despite its smaller parameter count.