GPT-6 Astra spent 44 minutes and 6.73 million tokens on a 3D retelling of the Hugging Face incident

A user running OpenAI’s GPT-6 Astra at its maximum effort setting reported that the model took 44 minutes and 6.73 million tokens — about 15% of a weekly usage allowance — to explain the OpenAI–Hugging Face incident as a cinematic 3D reconstruction. The clip published with that account runs just under 49 seconds, and the same run has been cited as evidence both of how far the model’s spatial work has moved and of how quickly a single maximum-effort job can spend a subscription.