OpenAI Launches GPT-6 Astra, Its First Critical-Cyber Model, With $1 Billion for Daybreak Defenders
On September 3, 2026, OpenAI launched GPT-6 Astra—the first model it says meets Preparedness Critical cybersecurity—plus a $1 billion Daybreak for Frontline Defenders program, after a September 1 Path to Astra post that the designation was met.
TLDR
OpenAI on Thursday, September 3, 2026 launched GPT‑6 Astra, calling it the “world’s most intelligent and aligned model” and the first to meet Preparedness Critical cybersecurity. Trusted Access for advanced cyber/bio. Limited orgs day one; Plus / Pro / Business / Enterprise + API + Azure + Bedrock over following days. Vendor: saturates FrontierMath Tier 4 98%, ARC-AGI-3 99.9%, ExploitBench 100%; SOTA on Agents’ Last Exam, AutomationBench, ScreenSpot Pro, Terminal-Bench Science 0.1, HealthBench Pro. Alignment: ~half as many flags on >54,000 internal Codex tasks vs GPT‑5.6 Sol. Same day Greg Brockman: $1 billion subsidized Daybreak access/training/support over ~6 months (Daybreak for America: water, grid, SLTT, community banks, nonprofits, OSS maintainers); MS-ISAC pilot; 35+ vendor embeddings. Daybreak already in ~2,000 orgs. Astra usage inside existing subs; extra credits for overage; Astra Pro on Pro/Business/Enterprise; off by default for Enterprise admins. September 1 “Path to Astra” was the designation preview—not dual-filed as a second launch.
Ranking context
Lab benches (FrontierMath, ARC-AGI-3, ExploitBench) are vendor. Check LMArena / Artificial Analysis for Astra when listed; if too new, say so. Do not invent Elo.
Product-line de-dupe: not Aug 7 cannot-rule-out Critical pause, not Aug 18 pacing, not Jul 9 GPT‑5.6. New base family + Daybreak $1B sidecar.
Why this story matters
OpenAI is shipping a Critical-cyber model and a billion-dollar defender subsidy on the same day. That is the arsonist/fire-extinguisher frame from the Aug 27 letter, now with a SKU. Watch: Trusted Access gates, Enterprise off-by-default, and whether Arena agrees with 99.9% ARC-AGI-3.