MiniMax Ships M3.1 Flash: Agent Tasks Match Top Models at Nearly Triple the Speed and 70% Lower Cost
Synopsis
MiniMax has released M3.1 Flash, a new Flash-series model aimed at agentic tasks, coding, computer use and cybersecurity. The company says it reaches the level of leading models on complex agentic tasks, coding and computer operation benchmarks while making online inference nearly three times faster and cutting usage costs by 70%. The model is available through the MiniMax API, M Plan and MiniMax Code, with weights to be released later.
Interpretation
M3.1 Flash moves the Flash line from a low-cost option toward matching leading models on agentic tasks, coding, computer use and cybersecurity while keeping costs low. Earlier Flash models were defined by efficiency; M3.1 Flash is described as a generational leap on these tasks, with online inference nearly three times faster and usage costs 70% lower. The stated comparison is that on Automation-Bench, DeepSWE and Agents' Last Exam it performs on par with top models at very low task cost.
The model reports checkable scores across enterprise workflow automation, software engineering and real computer operation. These capabilities were previously spread across different models; M3.1 Flash aims to cover the path from understanding a task to acting and delivering results within one model. It ranks among top performers on AutomationBench-AA, scores 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam and 82% on CyberGym.
The inference gain comes from more aggressive sparsity, giving everyday output speeds of 200+ TPS with median speeds consistently at 170+ TPS. Inference speed and cost were the main constraints of the Flash line, and the sparsity adjustment acts directly on both. Listed pricing is $0.15 per 1M input tokens, $0.60 output and $0.015 cached input, with overall agentic-task usage cost 70% below its predecessor.
Perspective
The report covers what MiniMax's release announcement states: M3.1 Flash is available through the MiniMax API, M Plan and MiniMax Code, and model weights will be released later. Benchmark scores and cost figures come from the company's own account and are relevant to readers tracking agent deployment cost and inference speed. The company also says a Pro version, aimed at pushing the boundaries of intelligence, will arrive later.
MiniMax says M3.1 Flash reaches the level of leading models on several benchmarks but gives no item-by-item comparison with specific models. The in-depth cybersecurity evaluation was performed by internal security experts, making it an internal assessment. Weights are not yet released, so third-party replication and independent evaluation remain to be seen.