Recommend using claude code with domestic models
Categories:
I actually have some aversion to the company anthropic, but it is currently at the height of its popularity, so I have to hold my nose and use it. I subscribed to a claudecode pro plan, but the quota is very small and can only be used for planning. Domestic models perform slightly worse, but the quota is more, offering clear cost-performance advantages. Domestic model providers currently do not offer a response API, so they cannot be well integrated into codex, provided we do not consider the old version of codex. Chat completion may have its historical limitations, causing the API performance of each provider to not be completely consistent. For example, deepseekv4 requires extra body to enable thinking and only recognizes “high/max”; multi-turn conversations with tools need to retain reasoning_content. Zhipu’s temperature cannot be set to 0. Under kimi’s thinking mode, temperature must be fixed at 1, tool_choice=required is not supported, and functions are not supported, etc. In other words, the openai API chat completion has not perfectly become the de facto standard. Direct integration will involve compromises in some scenarios, and in severe cases it will throw errors and fail to run continuously. On the other hand, regarding their compatibility with the anthropic API, although incompatibilities still exist in some scenarios, they are still making every effort to maintain compatibility. Therefore, using the anthropic API directly is a worry-free choice. And as an official anthropic product, claudecode has absolute say over anthropic’s API changes; if model providers do not want to lose claudecode users, they can only do their best to adapt. This is a suitable and worry-free choice for now. I hope that in the future a more universal API will emerge to completely decouple models from tools.
