DEDUPEDMiniMax released M3.1 Flash on its API at $0.15 per million input tokens. It is described as a large mixture-of-experts model aimed at agent work, with near-frontier agent scores at Flash-tier cost.
MiniMax also confirmed that Space Bunny, an anonymous model that processed more than 200 trillion tokens in stealth, was an early version of this release.
Another cheap Chinese model arriving the same day the US moved against DeepSeek shows how fast low-price competition keeps coming.
Our take
Price is the feature. Agents burn tokens, so cost per million decides which model wins.
From the issue
The US just banned DeepSeek from federal agencies, two days after its best model yet Monday, October 19, 2026