What is the expected behavior for hitting max output tokens? BAML just returns this error

Stream ended prematurely without completion marker (likely timeout)

but the stream iterator does not seem to finish gracefully. looking at the specific HTTP response on the dashboard i see the last stream delta was:

{
  "type": "message_delta",
  "delta": {
    "stop_reason": "max_tokens",
    "stop_sequence": null
  },
  "usage": {
    "input_tokens": 3,
    "cache_creation_input_tokens": 120,
    "cache_read_input_tokens": 68184,
    "output_tokens": 8192
  }
}

Hmm, that’s interesting… that delta format is surprising to me. This seems like an edge case we missed. We’ll look into this soon, thanks for flagging!

What version of baml are you on? We did patch something like this

We are on 0.214.0 currently

on 0.216.0 https://github.com/BoundaryML/baml/blob/canary/CHANGELOG.md#02160---2025-12-31 i addressed a timeout issue

Ok thanks ill try upgrading. What is the expected behavior in this case?

no timeout, but there should be a FinishReasonError. I will take a look at what we do here

yeah curious if the iterator is intended to return cleanly in this case or throw

actually dpeends on your finish_reason_allow_list https://docs.boundaryml.com/ref/llm-client-providers/anthropic#baml-specific-request-options

cool ok