POST /v1/chat/completions

OpenAI-compatible chat completions. Same request shape, same SSE streaming contract.

POST https://api.refuseless.dev/v1/chat/completions

Request

Field	Type	Required	Notes
`model`	string	Yes	Catalog id, e.g. `glm-5.3-refuseless`
`messages`	array	Yes	OpenAI chat messages
`max_tokens` / `max_completion_tokens`	number	No	Default `1024`. Clamped to your credit reservation.
`stream`	boolean	No	Default `false`. `true` returns SSE.
`tools`	array	No	OpenAI function-calling tools. Counted toward input reservation.
`tool_choice`	object/string	No	Passed through
`response_format`	object	No	Passed through
stream extras	—	No	`stream_options: {"include_usage": true}` supported

prompt_tokens on the request is ignored — the server re-estimates.

Non-streamed response

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "created": 1700000000,
  "model": "glm-5.3-refuseless",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "…",
        "tool_calls": [],
        "reasoning_content": "…"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 64,
    "total_tokens": 76
  }
}

content is null when the turn is tool-calls only (OpenAI behavior).
tool_calls appears with finish_reason: tool_calls.
reasoning_content appears on reasoning models (also split out of <think>…</think> content) — see Reasoning.
usage.prompt_tokens_details.cached_tokens appears when prompt caching applies.

Example

curl https://api.refuseless.dev/v1/chat/completions \
  -H "Authorization: Bearer $REFUSELESS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-refuseless",
    "messages": [{"role": "user", "content": "Write a tight system prompt."}],
    "max_tokens": 512
  }'

Errors

400 missing model · 401 bad key · 402 out of credits · 404 unknown model · 429 model busy / warming (honor retry-after) · 502/503 upstream or gateway trouble. Full table: Errors.