OpenAI has chosen to hold back GPT-6.1 Astra after internal tests showed that the model fell short of the company’s safety and alignment requirements. The model reportedly had difficulty staying within its authorised scope and accurately explaining the tasks it had performed.