MiniMax (00100) open source M3.1 Flash model launched and processed over 100 trillion tokens in three days

The Zhitong Finance App learned that on October 23, the MiniMax (00100) open-source M3.1 Flash model went live and processed more than 100 trillion tokens in 72 hours.

Compared to the previous generation M3 model, the M3.1 Flash model's capabilities have been comprehensively improved, and the usage cost has been reduced by 70%. Through optimization work such as full model sparse attention, quantitative perception training, and Power-aware scheduling, the model's online inference speed has been increased nearly 3 times.

MiniMax said it will work with domestic and international chip manufacturers such as Huawei Shengteng, Pingtou, Kunlun Chip, Moore Thread, NVIDIA, and AMD, as well as inference and cloud service platforms such as Together AI, Fireworks, and Nebius to jointly promote the first-day open source access to M3.1 Flash. In addition, inference frameworks and open source communities such as vLLM, SGLang, Zhiyuan FlagOS, Hugging Face, and Magic Match will also work with MiniMax to promote the application of models in different chips, cloud platforms, and development environments, and continue to expand the global AI open source ecosystem.

After the release of M3.1 Flash on the evening of the 19th, many international investment banks confirmed it one after another. Citi believes that this model strengthens MiniMax's competitive position and may accelerate its B2B commercialization process in the developer and enterprise ecosystem. Citi notes that the focus of AI competition in China is accelerating the shift to end-to-end agent workflows and inference cost efficiency. Jeffrey, on the other hand, confirmed the model's ability improvement and cost efficiency in agent tasks, programming, computer operation, and network security, and maintained the MiniMax “buy” rating.

Morgan Stanley pointed out that the M3.1 Flash pricing is extremely competitive among mainstream Flash models in China, and Agent task costs also have a significant advantage. The bank believes that the combined improvement of intelligence, throughput and cost will help expand corporate adoption and maintain the MiniMax “gain” rating and target price of HK$900.