MiniMax M3.1 Flash Cuts AI Agent Costs 70 Percent, Lifts Inference Speed Threefold
MiniMax says the upgrade lowers agent task costs by 70 percent and raises inference speed roughly threefold while claiming the top score among Chinese-made AI models on OSWorld V2 and Agents' Last Exam, so the central question for buyers is whether cheaper, faster runs outweigh the gap to the previous model on efficiency rather than raw capability.
Reporting from 1 source: GIGAZINE.
MiniMax announced MiniMax M3.1 Flash on October 19, 2026. It reports the model scores 76.2 on BankerToolBench, 73.7 percent on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82 percent on CyberGym, where MiniMax claims the model ranked first among Chinese-made AI models. Output speed exceeds 200 tokens per second, inference runs about three times faster than the previous model, and agent task costs fall 70 percent. Pricing is $0.15 per million input tokens and $0.60 per million output tokens, with cached input at $0.015 per million.
MiniMax says an early version of M3.1 Flash ran anonymously as Space Bunny and processed more than 200 trillion tokens before the model was announced. The launch pricing undercuts what the company says the previous model cost to run: $0.15 per 1 million input tokens, $0.60 per 1 million output tokens, and $0.015 per 1 million cached input tokens. MiniMax frames the gains as computational efficiency rather than scaling, and presents burst output speed above 200 tokens per second with a median of more than 170.
Independent evaluation provides the calibration. Artificial Analysis scored M3.1 Flash 42 on its Intelligence Index, against 29 for the previous MiniMax M3, and 53 percent on SciCode and 48 percent on Humanity's Last Exam. The same evaluator puts the model 4th across the 173 it tracks and well clear of the 13 median for models at the same price.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.