CHINA DROPS A BOMB ON WESTERN AI — MINIMAX LAUNCHES OPEN-WEIGHT MODEL THAT CLAIMS TO BEAT DEEPSEEK AT CODING Shanghai-based MiniMax launched its M3 foundation model on June 1, positioning it as the first open-weight AI system to combine frontier-level coding performance, a one-million-token context window, and native multimodal capabilities in a single package. The company claims M3 tops the open-weight SWE-Bench Pro coding benchmark at 59.0%, edging out competitor Kimi K2.6 and outpacing DeepSeek on multiple metrics. If the benchmarks hold under independent scrutiny, M3 represents a significant escalation in China’s AI capabilities and adds another name to the growing list of Chinese models cutting into the market share of Western AI incumbents. DeepSeek’s V4 Pro still leads on LiveCodeBench and Codeforces, but MiniMax is narrowing the gap in ways that should put OpenAI and Anthropic on notice. The model being open-weight matters enormously. Anyone can download and deploy it, which means its capabilities are not locked behind a paywall or an API. That dramatically lowers the barrier for companies, governments, and individuals who want access to frontier-level AI without paying Western prices or accepting Western terms of service. MiniMax is also preparing for a domestic IPO in China, signaling the company intends to compete seriously and for the long term rather than simply grab headlines with benchmark scores. Keywords: MiniMax M3, Chinese AI model, open-weight AI, AI competition China 2026, DeepSeek