Stream termination - Nito

The Sentinel

data: [DONE]\n\n

[DONE] is not JSON. Every stream ends this way, including one that carried an error chunk.

The Usage Chunk

Set stream_options.include_usage: true and one extra chunk arrives before [DONE], with an empty choices array and a usage object:

{
  "id": "...",
  "object": "chat.completion.chunk",
  "created": 1751299200,
  "model": "z-ai/glm-5.2",
  "choices": [],
  "usage": {
    "prompt_tokens": 24,
    "completion_tokens": 17,
    "total_tokens": 41
  }
}

A reader that indexes choices[0] unconditionally fails here; check for an empty array. The usage fields match the non-streaming response.

Order

The terminal content chunk (empty delta, non-null finish_reason).
The usage chunk, only if stream_options.include_usage was set.
data: [DONE].