MiniMax rolls out M3.1 Flash at $0.15 / $0.60: the Space Bunny stealth model was an early version

MiniMax officially rolled out M3.1 Flash on October 19 and named it as the model behind Space Bunny, the anonymous stealth model that appeared free on OpenRouter and OpenCode on September 23. MiniMax's launch post says Space Bunny was an early version that processed more than 200 trillion tokens while it ran anonymously. MiniMax teased the reveal on X at 09:21 UTC, and the launch post went up later the same day.

THE PRICE

$0.15 per million input tokens, $0.60 per million output and $0.015 for cached input, at a flat rate, per MiniMax's post. That is half of M3's current rate of $0.30 / $1.20 for prompts up to 512K tokens, per MiniMax's pricing page, which did not list M3.1 Flash yet when we read it. MiniMax says agentic tasks cost 70% less overall than on its predecessor.

WHERE TO GET IT

The MiniMax API, MiniMax's M Plan subscription and its MiniMax Code app. The models.dev catalog lists MiniMax-M3.1-Flash on MiniMax's API with a 1M-token context window and up to 512K output tokens. The weights are not out yet: MiniMax says they "will be released later."

WHAT MINIMAX CLAIMS

DeepSWE v1.1 73.7%, OSWorld V2 61.7, Agents' Last Exam 29.6, CyberGym 82% and BankerToolBench 76.2. MiniMax says its OSWorld V2 and Agents' Last Exam scores are the best of any Chinese model, and that output runs at more than 200 tokens per second, with inference nearly three times faster. These are MiniMax's own numbers; no independent board has scored the model yet. For scale, DeepSeek's own figures for its V4.1 Flash 1017 build are 77.0 on DeepSWE v1.1 and 35.4 on Agents' Last Exam, at the same $0.15 / $0.60 off-peak price.

WHAT COMES NEXT

MiniMax says a Pro version "will also arrive later." We are now watching for it.

WHAT WE DO NOT KNOW

When the weights and M3.1 Pro will ship, and how Artificial Analysis, Vals and the other boards will score M3.1 Flash.