Chinese AI company MiniMax on October 19, 2026 formally rolled out MiniMax M3.1 Flash, a new entry in its Flash line aimed at agentic work, coding, computer use, and cybersecurity. The company says the model makes online inference nearly three times faster than its predecessor while cutting overall usage costs for agentic tasks by 70%.
MiniMax also cleared up a recent mystery. The company says “Space Bunny,” a model that had been running anonymously and processed more than 200 trillion tokens, was an early version of M3.1 Flash.
Pricing is flat: $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens. MiniMax’s API documentation lists the model as MiniMax-M3.1-Flash with a 1 million-token context window covering input and output combined, multimodal input spanning text, images, and video, and an adjustable “effort” setting for thinking depth with five levels from low to max, defaulting to max.
On benchmarks, MiniMax reports 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 company says the OSWorld V2 and Agents’ Last Exam results rank first among Chinese models. These are MiniMax’s own figures; independent evaluations had not been published at the time of writing.
MiniMax attributes the speedup to more aggressive sparsity, saying output can exceed 200 tokens per second in everyday use, with median speeds consistently above 170 tokens per second.
M3.1 Flash is available now through the MiniMax API, the company’s M Plan subscription, and its MiniMax Code desktop app. MiniMax says model weights will be released later and that a higher-end Pro model is also on the way, without giving dates in its announcement.
DigiEditorial treats MiniMax’s October 19 announcement and API documentation as the primary sources, and corroborated the launch, speed, and cost claims against Reuters’ October 19 report and coverage by China’s Yicai.