Xiaomi’s MiMo-V3-Pro benchmark results have reportedly leaked, scoring 72.8 on SWE-Bench Pro, approaching top-tier overseas closed-source models.

Max For AI has published a disclosure noting that a benchmark screenshot allegedly of Xiaomi’s next-generation large model MiMo-V3-Pro has circulated in online communities. The screenshot shows the model is primarily targeted at coding agent and general agent scenarios, with some test scores approaching those of top overseas models like Claude Opus and GPT. According to the alleged screenshot, MiMo-V3-Pro scored 72.8 points on SWE-Bench Pro, higher than GLM 5.3’s 67.9 points and Kimi K3’s 65.8 points, and less than 3 points behind Claude Opus 5’s 74.6 points and GPT-5.6 Sol Max’s 75.4 points. It also scored 70.6 points on Terminal-Bench 2.0, similarly close to Claude Opus 5’s 72.0 points and GPT-5.6 Sol Max’s 73.5 points. Additionally, its τ3-bench score reached 76.4 points, compared to GPT-5.6 Sol Max’s 78.8 points. If these scores are ultimately officially confirmed and reproducible in the official version, MiMo-V3-Pro could enter the first tier of the world’s top models. However, the authenticity of the benchmark screenshot and its test conditions have not been confirmed by Xiaomi’s official, and the relevant data should still be considered unsubstantiated leaks. It is worth noting that Xiaomi’s latest officially announced MiMo flagship models are currently MiMo-V2-Pro and MiMo-V2.5-Pro, so whether MiMo-V3-Pro exists and its specific release date remain to be further disclosed by the company.