The Kimi K3 API is designed to be used through Moonshot's OpenAI-compatible interface. For most developers, that means the shape feels familiar: a base URL, an API key, a model name, and a chat messages array.
Use the OpenAI-compatible base URL
Kimi's public API documentation points developers to the Moonshot API and its OpenAI-compatible chat interface. The important pieces to verify before debugging your application are the base URL, the authorization header, and the model name.
For Kimi K3, use the model name documented by Kimi for the endpoint you are calling. If a gateway or tool rewrites model names, confirm the final request reaching Kimi still targets Kimi K3.
Shape your messages deliberately
The chat API accepts a conversation-style messages payload. Keep system instructions stable when possible, because stable prefixes are easier to cache and audit.
For coding or research workloads, do not send an entire repository or document set by default. Start with the minimum context needed for the task, then add more context only when the model needs it.
Plan parameters before production
Parameters such as temperature, max output, and tool usage affect both quality and cost. A production integration should log token counts, request IDs, errors, and retry causes.
Once you have logs, use them to update the calculator rather than relying on estimates from a single test prompt.