About OpenClaw's domestic large models: My lobster has been using two models: one is GPT Codex 5.3, and the other is Kimi K2.5. Now, I've switched Kimi to @XiaomiMiMo's Mimo V2 Pro. > What broke my defenses with Kimi was the first image. I asked it to help me check out the new version of OpenClaw and tell me the important update info. But it directly upgraded my system, which I found ridiculous because I gave a very clear instruction that didn’t request an upgrade, and it didn’t even give me a second confirmation. After using Kimi on and off for over two months, the overall experience has been decent. But one strong impression is that Kimi tends to talk a lot—like that employee who gives you tons of responses and feedback—but there’s also a lot of noise. You have to spend time extracting the core information yourself. > On the other hand, after using Mimo for a while, I feel it’s much cleaner and more efficient. I gave it a piece of text and asked the Agent to help me find background info. Compared to Codex, Mimo focused on some key points in the information I provided. Codex, however, returned more generic and vague responses. Mimo occasionally cracks jokes with me, but it doesn’t feel annoying. I think its handling of the Chinese context is much better, more native, and gives a greater sense of reliability. I believe Xiaomi has its unique advantages in the field of large models, especially with its ecosystem of people, cars, and homes. I’m really looking forward to seeing what kind of capabilities large models can unleash with Xiaomi’s massive device connectivity.