Xiaomi's MiMo-V2.6-Pro, released and open-sourced on September 22 under an MIT license, now leads open-weight large language models on the Artificial Analysis Intelligence Index with a 46 score, surpassing rivals like Moonshot's Kimi K3 and Alibaba's Qwen series while adding native multimodal support and a 1M-token context window. The gains stem from scaled reinforcement learning completed in under six days at roughly $2.6 million, paired with aggressive API pricing that has driven leading usage shares on platforms like OpenRouter. This rapid capability jump and distribution edge underpin the 94.4% market-implied odds. A late-September release from another lab posting higher benchmark results or credible questions around evaluation methodology could still shift the outcome before resolution.