Worked on Cypher’s model controls today. The old selector showed model names, but not reasoning levels. That made it incomplete. Picking the model matters, but so does knowing how hard it is being told to think. So I split the controls. Model selection and reasoning effort are now separate. GPT-5.5 with medium reasoning is the default. GPT-5.4 mini with low reasoning is the fallback if the selected setup fails because of access, availability, or compatibility. Then I tested it against a real response instead of assuming it worked. Requested: GPT-5.5 Used: GPT-5.5 Reasoning: medium Fallback: false. That is the result I wanted.