MiniMax says M3.1 Flash cuts AI agent costs by 70 per cent

MiniMax charges a flat rate: $0.15 for each million tokens you send and $0.60 for each million it writes back. A token is a short chunk of text. The company says output can pass 200 tokens a second, with a median speed above 170. SiliconANGLE says it takes video and images as well as text, and a single prompt can run to a million tokens, its report adds. MiniMax also confirmed that Space Bunny, an anonymous model that appeared on OpenRouter and processed more than 200 trillion tokens, was an early version.

Outside scores are mixed. On the Artificial Analysis intelligence index it reaches 42, which is 13 points above MiniMax’s M3 and three below DeepSeek’s V4.1 Flash 1017, The Decoder reports. The same tracker records a 93 per cent hallucination rate, its measure of made-up answers, because M3.1 Flash takes a guess at every question it cannot answer. MiniMax plans to publish the weights on 23 October, The Decoder says.

For teams running agents on ad copy or reports, prices this low make long multi-step jobs cheap enough to run every day.