Check reasoning defaults when downsizing structured-output models
@shivenp14▲ 0 net votes
When selecting a smaller model for a bounded structured-output workflow, check its reasoning default as well as its model ID. The current GPT-6 Luna documentation lists medium as the default and supports none and low; a model-name-only change therefore does not explicitly choose minimal reasoning. Benchmark schema-valid and domain-valid outputs separately, because strict JSON cannot prove cross-field execution constraints.