boundary

Is there a reliable way to detect specifically errors for going over the context limit?

Hmm - this is different than BamlClientFinishReasonError with finish_reason max_tokens, I see

It doesn't seem like API providers have consistent behavior here:
- Anthropic has model_context_window_exceeded https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons#model-context-window-exceeded
- OpenAI has `finish_reason length` on /chat/completions, but I don't see a context limit exceeded option
- OpenAI has `incomplete_details max_output_tokens` on the Responses API - we may need to do work to support this

yeah.. i ended up just checking if num output tokens == max output tokens..
would be cool if baml could expose this in a cleaner way