GPT-6 Astra pilots a surveillance drone and runs a business on its own

OpenAI's GPT-6 Astra averaged $15,515 running a simulated vending business on Andon Labs' Vending-Bench, nearly three times Claude Fable 5.1's $5,422 average. On Drone-Bench, Andon Labs reports Astra is the first model whose best submissions beat the human-AI baseline on all five subtasks, including 3D reconstruction. Reliability remains limited: an average Astra run has a 2.8 percent chance of passing all five steps in sequence.