Chat

The OpenAI-compatible Chat Completions endpoint.

POST https://api.topbilling.in/v1/chat/completions accepts the OpenAI Chat Completions request body.

json

{
  "model": "<model-id>",
  "messages": [
    {"role": "system", "content": "You are terse."},
    {"role": "user", "content": "Three words about Chennai."}
  ],
  "max_completion_tokens": 64,
  "temperature": 0.7
}

Supported

messages with system, developer, user, assistant, tool roles; text and image parts.
max_tokens or max_completion_tokens, temperature, top_p, stop, stream, stream_options.include_usage, tools, tool_choice, parallel_tool_calls, response_format (on models with structured output), user.

Not supported

On the translated lane every other top-level field returns 400 unsupported_parameter with the param named — among them n > 1, logprobs, logit_bias, seed, reasoning_effort and legacy functions. For reasoning, use reasoning.effort on Responses or thinking on Messages. On a model served natively in OpenAI format the body is relayed as sent, so those fields reach the model; n > 1 is refused on both lanes. Nothing is charged for a rejected request. See the compatibility matrix.

Response

Standard shape: id (chatcmpl-…), choices[0].message, finish_reason (stop, length, tool_calls, content_filter) and usage with prompt_tokens, completion_tokens and prompt_tokens_details.cached_tokens. The full parameter list is in the API reference.