GM☕️ Xiaomi quietly launched the MiMo V2 Pro last night, which was previously speculated to be the mysterious model DeepSeek V4, but it turned out to be from Lei Jun's family.🤣

A few key numbers: 1T parameters, 42B activations, 1 million tokens of context. Scored eighth globally and second domestically, with API pricing around 15% of GPT-5.2. Free to use for a week starting today.

Has anyone tried it yet? I'm really curious about how well it works, waiting for feedback online.

The 1 million tokens of context makes me think of lobsters—AI agents rely heavily on context length to work effectively, which has been a bottleneck until now. Xiaomi now directly provides 1M, feeling like the lobsters can feast more.

Speaking of mobile phone manufacturers creating large models, it's quite magical. Lei Jun has transitioned from "born for enthusiasts" to "born for inference," is the next step to run models in the Xiaomi SU7?🤣