MiniMax M3.1 Flash: cheaper, faster agentic coding model, weights Oct 23

MiniMax has revealed that "Space Bunny", an unnamed model that had been tested publicly as a stealth model in September, was an earlier version of its new M3.1 Flash (x.com). The company says Space Bunny topped global leaderboards with more than 200 trillion tokens, without naming which leaderboards. M3.1 Flash is pitched as the economical sibling of MiniMax-M3: MiniMax claims frontier-level results on agentic office work, coding and visual reasoning at a third of M3's cost and three times its speed, citing its own scores of 74% on DeepSWE, 62% on Automation-Bench, 30% on ALE and 91% on BabyVision. Those figures come from MiniMax's announcement; the thread does not include independent results.

API pricing is a flat $0.15 per million input tokens and $0.60 per million output tokens, with cached input at $0.015 per million, and MiniMax says the model weights will be released on October 23. MiniMax's model documentation lists M3.1 Flash next to M3 as a multimodal coding model with a 1M-token context window, adding tunable thinking depth as the feature M3's entry lacks (platform.minimax.io). A preview of M3.1 Flash had already been offered to subscribers of MiniMax's Token Plan in late September.