Morgan Stanley has issued a research report maintaining its "Overweight" rating on MiniMax-W (00100) with a target price of HK$900.
The bank noted that MiniMax officially launched the M3.1 Flash model on Monday (October 19), and has not yet disclosed the model's parameters or licensing information. The model shows significant improvements in agentic workflows, coding, computer use, and cybersecurity, achieving a DeepSWE score of 73.7%, an OSWorld V2 score of 61.7, and a CyberGym score of 82%, with OSWorld V2 and CyberGym ranking among the top models in China.
The bank pointed out that although M3.1 Flash's pricing is not the lowest, it is competitive among major Flash large language models in China, and its agentic task cost is also advantageous. By optimizing the sparse attention mechanism, the model's inference speed has increased by nearly 3 times, while agentic task costs have been reduced by 70%; in terms of API pricing, each million input and output tokens are priced at RMB 1 and RMB 4, respectively.
In the Artificial Analysis Intelligence Index, M3.1 Flash ranks second among major Flash large language models in China, behind only DeepSeek V4.1 Flash Oct. The bank believes that the combination of higher intelligence, higher throughput, and lower cost should help expand enterprise adoption.
The bank expects that M3 Pro will become the primary large language model demonstrating MiniMax's pursuit of frontier intelligence capabilities, and is expected to be launched by the end of October.
Comments