MiniMax announced the M3.1 Flash model on Oct 19, 2026. The company says the new version scores 42 points on the Artificial Analysis Intelligence Index, 13 points higher than its previous M3 model, while costing less than half as much per task. The API is already live at platform.minimax.io and the company plans to make the weights open on Oct 23, 2026.
How the model performs
In tests, M3.1 Flash reached 48 % on the “Humanity’s Last Exam” benchmark, beating Deepseek V4.1 Flash’s 43 % and topping GLM‑5.3‑Flash. On the physics‑heavy CritPt benchmark it scored 22 %, ahead of GLM‑5.3‑Flash (15 %) and Gemini 3.8 Flash (18 %) but just below Deepseek’s 24 %. For factual accuracy on the AA‑Omniscience test the model answered 56 % of questions correctly, well above GLM‑5.3‑Flash (28 %) and Deepseek (43 %). The model can handle up to one million tokens of context and accepts both text and images.
Pricing and availability
MiniMax charges $0.15 per million input tokens and $0.60 per million output tokens, with cached inputs at $0.015. Because M3.1 Flash uses fewer output tokens, the cost per task is about $0.22, roughly three‑times faster than the older M3. The company also offers a desktop app called MiniMax Code.
Why it matters
For developers and businesses in China and elsewhere, the lower price per task means they can run more complex reasoning jobs without breaking the budget. The open‑weight release could also let researchers experiment with a high‑performing model without waiting for a commercial license.
