Spending a lot of money using Claude Sonnet with OpenClaw for on chain data analysis may be one of the biggest mistakes of the past two years. CZ just publicly stated that he has tested several models, and Kimi is the most token efficient among them. Recommend several cheap models suitable for using OpenClaw for crypto tasks: 1. Kimi (Dark Side of the Moon) - Long context has high efficiency, recommended by CZ for personal testing, and consumes significantly less token than Claude. 2. DeepSeek V3- Chinese crypto report scenario is very stable, with prices much lower than the Claude series, making it the preferred choice for high-frequency batch queries. 3. Gemini 1.5 Flash - produced by Google, with free quota, fast speed for lightweight query tasks, and the most cost-effective when heavy analysis is not required. 4. Qwen Plus has advantages in ultra long context scenarios, as it is cheaper than other models when analyzing large amounts of on chain data. Switching method: OpenClaw Settings → Model Configuration → Select Model → Fill in the corresponding API key, 5 minutes to complete. I am currently using the first and second to run daily on chain analysis, while Claude is keeping it for tasks that require complex reasoning, resulting in significantly fewer bills compared to before. To write code, call the local claude cli.