I feel that Chinese blogs have little hope left, with very few readers and traffic from search engines declining day by day. However, some programming-related content actually gets more traffic from overseas, even in Chinese. So, I thought about automatically translating my Chinese blog into English. This might reach a larger audience.
Previously, I tried Free AI Large Model API Interface: Golang Proxy Implementation for Gemini 3 Flash Preview Version. Even the smaller model performed well, but accessing Google from domestic servers is a major hassle. Although I could access it through a proxy in Germany, after much hesitation, I abandoned this solution. One reason is the low monthly quota for Gemini-related models, and another is the reluctance to maintain another proxy service.
Using the Github Copilot SDK for Translation Services
Refer to the previous article: Unlimited Free Large Model Tokens: Github Copilot CLI SDK Installation and Testing. Since the Github Copilot SDK provides free large models like gpt-5-mini and gpt-4.1, both are more than sufficient for English translation.
I am unsure whether deploying a Copilot CLI on a server would lead to an account ban because using the same Copilot Pro account on multiple machines with different IPs simultaneously always feels risky. Therefore, I adopted a solution similar to OpenClaw:
- The translation service based on the Github Copilot Golang SDK runs on my local machine, i.e., my personal laptop.
- The translation service on this laptop connects to my server’s blog service via WebSocket.
- When a Chinese content in the blog needs translation, the translation task is automatically distributed through WebSocket.
- After the local translation service completes processing, it returns the result to the server. The server then publishes the new content.
The effect is quite good.
Model Selection
Both gpt-5-mini and gpt-4.1 work well. However, avoid using gpt-4o because actual tests show that gpt-4o consumes Premium requests quota. Moreover, each request consumes 1x… which is even more expensive than Gemini 3 Flash. That’s frustrating. But the model list in VSCode’s Copilot shows that gpt-4o is free. I was scared when I saw how much quota I consumed today.
I found a similar issue on GitHub:
https://github.com/github/copilot-sdk/issues/334
Some models are missing from CopilotClient.list_models. The issue indicates that several models (Gemini 3 Flash, Groke Code Fast 1, GPT-4o, and Raptor mini) are not appearing in CopilotClient.list_models()
Although it doesn’t mention the issue of consuming Premium requests, it seems that the Copilot SDK is a bit confusing.
About the Author 🌱
I am a developer from Yantai, Shandong, China. If you have any interesting topics or software development needs, feel free to email me at: zhongwei.sun2008@gmail.com for a chat, or follow my personal public account "Elephant Tools", See more contact information