MiniMax M3.1 Flash shows China's AI price race has no floor in sight
decoder ·
MiniMax has released its new reasoning model, M3.1 Flash, which demonstrates a significant reduction in cost while maintaining competitive performance in China's intense AI pricing race. The model scores 42 points on the Artificial Analysis Intelligence Index, marking a substantial improvement over its predecessor and placing it just behind Deepseek V4.1 Flash. It achieves this efficiency by using fewer output tokens, resulting in a task cost of approximately $0.22, which is lower than Deepseek's $0.26 per task. MiniMax leads in factual accuracy on the AA-Omniscience test with a 56 percent success rate, outperforming both Deepseek and GLM-5.3-Flash. However, the model suffers from a high hallucination rate of 93 percent because it attempts to answer every question rather than declining uncertain ones. Despite this weakness, it shows strong capabilities in complex reasoning benchmarks such as Humanity's Last Exam and CritPt. The API is currently live with specific token pricing, and MiniMax plans to release open weights for the model on October 23.
decoder