Chinese AI company MiniMax released MiniMax M3.1 Flash on October 19, 2026, featuring 428 billion total parameters, about 23 billion active parameters in a mixture-of-experts architecture, and a 1,000,000-token context window.