[Suspected Benchmark Leak of Xiaomi MiMo-V3-Pro, SWE-Bench Pro Score Reaches 72.8, Approaching Top Overseas Closed-Source Models] BlockBeats News, August 21: Max For AI published an article revealing that a suspected benchmark screenshot of Xiaomi's next-generation large model, MiMo-V3-Pro, has been circulating in the community. The screenshot indicates that this model focuses on Coding Agent and General Agent scenarios, with some test results nearing those of top overseas models like Claude Opus and GPT. According to the suspected screenshot, MiMo-V3-Pro achieved a score of 72.8 on SWE-Bench Pro, surpassing GLM 5.3's 67.9 and Kimi K3's 65.8, and coming within 3 points of Claude Opus 5's 74.6 and GPT-5.6 Sol Max's 75.4. On Terminal-Bench 2.0, it scored 70.6, similarly close to Claude Opus 5's 72.0 and GPT-5.6 Sol Max's 73.5. Additionally, its τ3-bench score reached 76.4, compared to GPT-5.6 Sol Max's 78.8. If these results are officially confirmed and reproducible in the final version, MiMo-V3-Pro could potentially enter the ranks of the world's top-tier models. However, the authenticity of the benchmark screenshot and testing conditions have yet to be verified by Xiaomi's official sources, and the related data should still be regarded as unconfirmed rumors. It is worth noting that Xiaomi's currently publicly disclosed flagship MiMo models are MiMo-V2-Pro and MiMo-V2.5-Pro, so the existence of V3-Pro and its specific release date remain subject to further official announcements. [Original Link]