How hard a reasoning model thinks. Not every model takes every value:
minimal is a GPT-5 setting that GPT-6 rejects, and none, xhigh and
max are GPT-6 settings. The OpenAI providers send low in place of
minimal to a GPT-6 model, so a configured minimal never fails a run.
How hard a reasoning model thinks. Not every model takes every value:
minimalis a GPT-5 setting that GPT-6 rejects, andnone,xhighandmaxare GPT-6 settings. The OpenAI providers sendlowin place ofminimalto a GPT-6 model, so a configuredminimalnever fails a run.