The Xiaomi MiMo team has officially announced the open-sourcing of its MiMo-V2.6 series of native full-modal models. On the same day, Elon Musk revealed the much-anticipated Grok 4.7 model. While both models achieved a score of 46 on the AA Comprehensive Intelligence Index, there is a notable disparity in their costs. The evaluation and operational expenses associated with the MiMo-V2.6-Pro are significantly lower than those of Grok 4.7, with a cost difference approaching 29 times. In terms of capability enhancement, Grok 4.7 places a greater emphasis on programming and knowledge-intensive tasks, boasting a 500K context window. Conversely, the MiMo-V2.6 model offers a wider array of capabilities, spanning multiple scenarios such as multi-agent development and 3D modeling, and features an expansive 1M context window. Throughout various tests, the MiMo-V2.6 outperformed Grok 4.7, demonstrating a stronger sense of completion in real-world generation tasks. The Xiaomi team disclosed that MiMo-V2.6 underwent large-scale reinforcement learning training via technical pathways such as MixRL and MOPD merging capabilities, incurring a total cost of approximately $3.47 million. This achievement represents a significant milestone in Xiaomi's pursuit of recursive self-improvement methodologies. Meanwhile, Musk has commenced the training process for Grok 4.8, which boasts a staggering parameter scale of 2.5 trillion.