MiniMax M3.1 Flash released: intelligence index up 13 points, ~$0.22 per task

MiniMax released reasoning model M3.1 Flash, which scores 42 on the Artificial Analysis Intelligence Index, 13 points above the prior M3, at less than half the per-task cost. It scores 48% on Humanity's Last Exam and 56% on the factual-accuracy test AA-Omniscience, both above GLM-5.3-Flash and Deepseek V4.1 Flash 1017; it scores 22% on research-grade physics reasoning benchmark CritPt, slightly below Deepseek's 24%. The model supports up to a one-million-token context, accepts text and image input, and is API-priced at $0.15 per million input tokens, $0.60 output, and $0.015 cached input. The cost advantage comes from fewer output tokens: about 189 million total output tokens in benchmarks versus Deepseek's 224 million, at roughly $0.22 versus $0.26 per task. It scores 73.7% on the DeepSWE v1.1 coding benchmark at under $0.30 per task. MiniMax says it runs three times faster than M3. Two groups are most affected: developers calling the MiniMax API must accept its 93% hallucination rate, since the model answers uncertain questions instead of declining, so they need to add their own validation or refusal layer; teams preparing to deploy open weights can get them from Hugging Face after October 23 for local or private-environment inference. Deepseek V4.1 Flash 1017 is already on Hugging Face, and the two trade wins on scores and price.