CHINA’S AI BLITZ: FOUR LABS DROP FRONTIER CODING MODELS IN 12 DAYS — AT A FRACTION OF WESTERN PRICES DeepSeek, Moonshot (Kimi), MiniMax, and Z.ai’s GLM all shipped powerful open-weight coding models within 12 days of each other in late April, with Kimi K2.6 briefly topping the SWE-Bench Pro coding benchmark above GPT-5.4 — at inference costs less than a third of Western rivals. The coordinated surge signals China’s AI labs are closing the capability gap faster than expected. Keywords: Chinese AI models 2026, DeepSeek, Kimi K2.6, AI coding benchmark, open weight models