12,456 characters of reasoning and 0 characters of answer came back from a 3000 token budget. I'd set reasoning to none. The client dropped the field instead of sending disabled, and the model treats a missing field as reasoning on. Enabled false gave me 996 tokens and an answer.