MiniMax says M3.1 Flash redefines the trade between capability and cost for agents, with inference nearly three times faster and usage cost 70% lower than its predecessor.
Space Bunny, which processed more than 200 trillion tokens while operating anonymously, was an early version of the model, the company said. The new Flash model is priced at $0.15 per 1M input tokens, $0.60 per 1M output tokens and $0.015 per 1M cached input tokens.
It is a short leap that covers a lot of benchmarks.
MiniMax said M3.1 Flash scores 76.2 on BankerToolBench for multi-step planning and financial compliance, 73.7% on DeepSWE v1.1 for autonomous coding, 61.7 on OSWorld V2 and 29.6 on Agents Last Exam for computer use, and 82% on CyberGym for vulnerability discovery. The company also said median output speed reaches 170+ TPS and can hit 200+ TPS in everyday use through more aggressive sparsity.
That pricing and speed pitch lands in a crowded Flash tier where the median matters. Per Artificial Analysis, MiniMax-M3.1-Flash is a reasoning model that generated 180M output tokens on the Intelligence Index, above the median of 99M for other models in a similar price tier. Per OpenRouter, the model is multimodal with a 1,000,000 token context window and a 524,288 token maximum output. MiniMax described top-tier parity on AutomationBench-AA, DeepSWE and Agents Last Exam at a very low task cost, but the company declined to detail the sparsity method or give a date for weight release beyond later.
