📌What model should be used for shrimp farming?丨Let's take a look at the performance of major AI large language models in OpenClaw proxy tasks——

Based on my current experience and some feedback collected from friends around me, here’s a brief overview of these models:
1⃣ Gemini: Currently recognized as one of the most stable Agent models, very reliable in task decomposition, tool invocation, and multi-step processes. The downside is that API load and rate limiting issues occur more frequently.
2⃣ MiniMax: Very low cost, high cost-performance ratio, but has a rather plain personality, making it suitable as a worker model in domestic automation pipelines.
3⃣ Kimi K2.5: Strong in long context and Agent architecture, ideal for document analysis and visual processing tasks, and also a domestically developed worker model. However, its stability is average, and it tends to write a lot of verbose code for complex processes, constantly fixing bugs.
4⃣ Claude Opus: The most reliable model for executing complex tasks, with extremely stable reasoning. The only downside is that it’s “expensive,” and costs can easily spiral, but mastering some tricks makes this manageable.
5⃣ GPT: Fast, suitable for small tools or lightweight tasks. It tends to forget context in complex tasks, and its instruction tracking stability is not as good as Gemini or Claude. Not very suitable for shrimp farming, so you can just pass on it.
In a nutshell—
Gemini: The most disciplined quant fund
Kimi: A highly talented but occasionally lazy trader
MiniMax: A low-paid but diligent intern
GPT: The most famous but volatile
Claude: Top-tier researcher
In conclusion, I chose Claude Opus + Xianyu — the absolute king of cost-performance!