MiniMax Releases M3.1 Flash Reasoning Model at $0.22 Per Task, Plans Open Weights

MiniMax released its M3.1 Flash reasoning model, scoring 42 points on the Artificial Analysis Intelligence Index—up 13 points from its M3 predecessor at less than half the per-task cost. The model ties with GLM-5.3-Flash, edges past Gemini 3.8 Flash (41), and trails DeepSeek V4.1 Flash 1017 (45). In API pricing, M3.1 Flash costs $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 for cached inputs. By generating about 189M output tokens across the benchmark compared to DeepSeek's 224M, it brings average cost per task down to about $0.22 versus DeepSeek's $0.26, and MiniMax says it runs three times faster than M3. On complex reasoning and accuracy benchmarks, M3.1 Flash hits 48% on "Humanity's Last Exam" and 56% on the AA-Omniscience factual test, outperforming DeepSeek V4.1 Flash 1017 (43% on both) and GLM-5.3-Flash. On the 71-problem CritPt research-level physics benchmark, it scores 22%, beating Gemini 3.8 Flash (18%) and GLM-5.3-Flash (15%) while trailing DeepSeek (24%). Artificial Analysis flagged a 93% hallucination rate as a weak spot, noting that the model attempts to answer every test question rather than declining when uncertain. The model supports a 1M-token context window for text and image inputs, is available via MiniMax's API and its MiniMax Code desktop app, and is scheduled for an open-weights release on October 23.