MiniMax launches M3.1 Flash model to reduce agentic workload costs

Chinese artificial intelligence developer MiniMax Group Inc. has released M3.1 Flash, an updated large language model designed to reduce the cost of running AI agents. SiliconANGLE, Reuters, MarkTechPost, and other outlets all reported on the launch of the new model. SiliconANGLE noted that M3.1 Flash is an update to the M3 model that MiniMax originally introduced in June. [1][2][3][4][5][6][7][8][9][10]

According to SiliconANGLE, MiniMax stated that running agentic tasks on M3.1 Flash costs 70 percent less than on its prior model. SiliconANGLE also reported that inference speed is close to three times as fast. MarkTechPost reported that the model uses a mixture-of-experts architecture with 428 billion total parameters, about 23 billion active parameters, and a context window of one million tokens. [1][5][7][8][9]

MarkTechPost reported that application programming interface pricing is set at 0.15 dollars per million input tokens and 0.60 dollars per million output tokens, with model weights scheduled for release on October 23. VentureBeat reported that M3.1 Flash outperforms DeepSeek and Opus 5 on a key business benchmark at lower cost, although it noted a high hallucination rate. As FTMQ SI reported earlier, competitor DeepSeek recently updated its own Flash model checkpoint to target developer costs. [5][10][13]