You can indeed download and bookmark this openclaw orange peel book. The author is @ AlchainRust Link: https://my. (feishu.cn)/wiki/H27Iw9ussiaYbokymhncExtjnAh To be honest, if you put this orange peel book in Xiaohongshu, you can earn some money for the course .. And regarding the model, let me share my own experience. Below are the models I have been using all along Claude sonnet4.6/opus 4.6 Currently, Claude's model is truly fault leading. In terms of daily conversation experience and code experience, it's really good. The only problem is that it's too expensive. If you want to go to a transfer station and are worried that your data will be distilled and sold, you may not be able to get a full blooded version. This is a rather awkward point Codex GPT5.4 This is mainly because I have been renewing ChatGPT, and I just remembered a few days ago that it can also be used. Overall, I feel that it is the most cost-effective option. What makes me feel the most friendly is that after installing openlcaw, many times you can ask Codex to help you check where the calls are problematic, and it can assist you with the operation. Very friendly to players with zero code foundation. Mainly because this thing is not expensive, 20U per month, very cost-effective KIMI K2.5 Before fully switching to GPT-5.4, I had been using the KIMI 199 one month package. How can I describe this model based on my experience? You say it's silly, there's no problem with daily conversation and communication, it feels quite intelligent; But once it comes to some deep programming, it's a bit insufficient. And sometimes I even tease you It's a bit like the mischievous child in the class who has average grades. He's not stupid at all, and he's not too smart either. He hands in his exams quite quickly, but his grades are not good. MINIMAX M2.5 I bought the version that costs $500 per year. I have been buying it for over a month, but I didn't use it until yesterday .. What is the main reason? It's because every time I ask KIMI to help me configure MINIMAX, it never gets configured properly, making me think I bought the wrong version Until a few days ago when I switched to CODEX, it immediately helped me fix it .. There is no significant difference in the specific experience yet, it remains to be observed And here are some tips I have tried claude max many times, and it seems that I have indeed logged in Oauth to BAN. Anyway, the MAX version I bought cannot be called normally every time. Even if packaged as SKILL, short texts can indeed be called once or twice, but long texts will get stuck in the reply I don't know how everyone solved this problem. If paying solely through APIs, it is indeed too expensive On the contrary, CODEX can be directly configured into Openclaw using Oauth, and the user experience is really good.