streaming terminal sentinel on OpenRouter APIs

[jonny64]

Docs page URL

https://openrouter.ai/docs/api-reference/streaming

API endpoint

/api/v1/chat/completions

Description

I observed a trailing [DONE] SSE sentinel on both OpenRouter streaming APIs:

• OpenAI-compatible: POST /api/v1/chat/completions
• Anthropic-compatible: POST /api/v1/messages?beta=true

Observed behavior

OpenAI-compatible path

Response tail:

data: {...final chunk...}
data: [DONE]

Anthropic-compatible path /api

Response tail:

event: message_stop
data: {"type":"message_stop"}

event: data
data: [DONE]

Repro models

Observed on:

• minimax/minimax-m2.5
• anthropic/claude-haiku-4.5

Docs question

I could not find explicit documentation that the Anthropic-compatible endpoint also emits a terminal [DONE] sentinel.

Is this intentional behavior for OpenRouter's streaming layer across both APIs?

If yes, could this be documented explicitly for /api/v1/messages as well?

Steps to reproduce

No response

[jonny64]

Impact: LiteLLM crash with

  Error in _route_streaming_logging_to_handler: Expecting value: line 1 column 3 (char 2)

Exact exception type/message from traceback:

  json.decoder.JSONDecodeError: Expecting value: line 1 column 3 (char 2)

Exact failing line from traceback:

  File ".../litellm/llms/anthropic/chat/handler.py", line 1153, in convert_str_chunk_to_generic_chunk
    data_json = json.loads(str_line[5:])

Exact upstream tail that triggered it:

  event: message_stop
  data: {"type":"message_stop"}

  event: data
  data: [DONE]

Workaround: jonny64/litellm@90bdeff would not be accepted in litellm upstream if undefined behaviour not covered, i think