GLM-5.3 spent the structured-call output budget on reasoning and got truncated (finish_reason=length) on every profile classification. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
1.0 KiB
1.0 KiB