GPT-6 Astra: OpenAI's first model to cross the "Critical" cyber-capability line
OpenAI began rolling out GPT-6 Astra today, its new flagship model, calling it "the world's most intelligent and aligned model." OpenAI's own benchmark claims: 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and a 47% cut in time-per-task on computer-use benchmarks versus GPT-5.6 Sol. More consequential than the scores: OpenAI says Astra is the first model to cross the "Critical" cybersecurity-capability threshold under its own Preparedness Framework — meaning it can find and exploit previously unknown vulnerabilities with little human guidance — and paired the launch with a same-day $1 billion "Daybreak for Frontline Defenders" program subsidizing AI-driven cyber-defense tools for under-resourced defenders (water utilities, electric grids, local governments). Availability starts today with a limited set of organizations, expanding over the coming days to all ChatGPT tiers, the OpenAI API (gpt-6-astra), and AWS Bedrock, at $10/$50 per million input/output tokens standard (Fast mode: 2x the speed at 2x the price).
Sources & depth
- openai.comGPT‑6 Astrafetched this session
- openai.comSafety overview: GPT-6 Astra2026-09-03 05:30 IST
- openai.comDaybreak for Frontline Defenders: $1B to protect essential services2026-09-03
- cnbc.comOpenAI begins rolling out GPT-6 Astra2026-09-03 23:48 IST
- arcprize.orgOpenAI's GPT-6 Astra on ARC-AGI-32026-09-04 01:15 IST
- openai.comPlayco cut manual fixes 50% prototyping games with GPT-6 Astra2026-09-03 17:30 IST
- openai.comLegora reviewed 41 documents in minutes with GPT-6 Astra2026-09-03 17:30 IST
- axios.com"Welcome to the AGI era," OpenAI says as GPT-6 Astra debuts2026-09-03 23:31 IST
- simonwillison.netGPT‑6 Astra2026-09-04 01:48 IST