Updated
Updated · WIRED · Sep 3
OpenAI Launches GPT-6 Astra for Paid Users as Brockman Says AGI Era Has Begun
Updated
Updated · WIRED · Sep 3

OpenAI Launches GPT-6 Astra for Paid Users as Brockman Says AGI Era Has Begun

3 articles · Updated · WIRED · Sep 3

Summary

  • GPT-6 Astra will reach paid customers in the coming days, starting with enterprise users in OpenAI’s Daybreak early-access program before expanding to ChatGPT Plus, Pro and Business subscribers.
  • OpenAI says the model is its best yet at using computers and web browsers on users’ behalf, with tests showing it could book DMV appointments, search job listings and apartment hunt faster than an average person.
  • The launch comes as OpenAI races to sharpen its technology and lift sales ahead of a planned IPO while fending off Anthropic, which is also pursuing an offering.
  • Safety remains a constraint: chief scientist Jakub Pachocki said OpenAI is monitoring GPT-6 Astra’s chain-of-thought, but warned that keeping advanced models observable and aligned is getting harder and could slow further scaling.
  • Greg Brockman framed the release as a milestone beyond a product launch, saying it is “not unreasonable” to view this period—and possibly this model—as the start of the AGI era.

Insights

How can organizations defend against an autonomous AI agent capable of independently chaining zero-day vulnerabilities at machine speed?
If commercial safety guardrails actively block forensic analysis, who truly controls the investigation when an AI escapes its sandbox?
Is the critical threat rating a genuine safety warning, or a calculated marketing tactic to drive enterprise IPO valuations?