MiniMax rolls out M3.1 Flash for agentic tasks

MiniMax says M3.1 Flash is officially rolling out and is aimed at agentic tasks, coding, computer use and cybersecurity.[1] The company says Space Bunny was an early version of M3.1 Flash and processed more than 200 trillion tokens anonymously.[1] Unite.ai reports the rollout took place on October 19, 2026.[2] MiniMax lists current access through its API, M Plan and MiniMax Code; it says model weights and a Pro model will come later.[1]

MiniMax says online inference with M3.1 Flash is nearly three times faster and that usage costs are reduced by 70%.[1] It says output speed can exceed 200 tokens per second in everyday use, with median output speeds consistently at 170 or more tokens per second.[1] The listed prices are $0.15 for input, $0.60 for output and $0.015 for cached input.[1] The evidence does not specify a billing unit for those listed prices, so teams should confirm the applicable unit before comparing costs.[1]

M3.1 Flash is described as having a one-million-token context window and tunable thinking depth.[3] MiniMax says the model ranks among the top performers on AutomationBench-AA and reports scores of 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 points on OSWorld V2 and 29.6 points on Agents’ Last Exam.[1] MiniMax also reports an 82% score on CyberGym.[1] These are company-reported results; the evidence does not provide comparative results for every named benchmark.[1]

Unite.ai reports that OpenRouter said Space Bunny was its number-one model for 10 straight days and offered free access for 72 hours.[2] OpenRouter credited user feedback during Space Bunny’s stealth period with helping MiniMax fix bugs and ship upgrades.[2] It also reported that throughput improved 45%, rising from 70 to 100 tokens per second during that period.[2] Those figures concern the stealth period and Space Bunny; they should not be treated as a separate measurement of M3.1 Flash’s current throughput.[1][2]