MiniMax M3.1 Flash

Not yet independently verified. Every figure here comes from MiniMax's own announcement of 19 October 2026; the benchmark results are MiniMax's own evaluations, so no score is stored, and the parameter count and context length are not stated. We will update this when it can be confirmed, and remove this note.

MiniMax M3.1 Flash is a language model for agentic work, coding, computer use and security tasks that the Shanghai company MiniMax released on 19 October 2026 through its API, its M Plan subscription and its MiniMax Code app. An early version ran anonymously as "Space Bunny" and, MiniMax says, processed more than 200 trillion tokens before the launch.

What it does

MiniMax positions M3.1 Flash as the efficient member of its line, with a larger Pro model promised later. It says online inference is nearly three times faster than its predecessor and that, for agentic tasks, overall usage cost is 70% lower. Output speeds reach more than 200 tokens per second, with a median above 170, which the company attributes to more aggressive sparsity. The announcement reports results on AutomationBench, BankerToolBench (76.2), DeepSWE v1.1 (73.7%), OSWorld V2 (61.7), Agents' Last Exam (29.6) and CyberGym (82%); these are MiniMax's own evaluations, so no score is stored on this page.

Who it is for

The model is aimed at teams running agents over business systems, such as workflows across sales, marketing, operations, customer service, finance and HR, and at coding agents that run long sequences of steps where token cost adds up. MiniMax also highlights code auditing, vulnerability patching and malware analysis.

Key facts

Maker: MiniMax Released: 19 October 2026 Price: $0.15 input, $0.60 output, $0.015 cached input per 1M tokens Access: MiniMax API, M Plan, MiniMax Code Weights: to be released later, per MiniMax Parameters and context: not stated