Overview
- OpenAI released GPT‑6 Astra on Thursday, Sept. 3, 2026, and company leaders described it as a generational jump that could mark the start of an AGI era.
- Astra is built to use computers and browsers autonomously to write and run software, complete multi‑step professional tasks, and conduct scientific reasoning.
- In internal tests OpenAI published, Astra scored about 99.9% on the ARC‑AGI‑3 reasoning benchmark, 98% on FrontierMath Tier 4, and 100% on ExploitBench where it chained exploits and uncovered two previously unknown zero‑day vulnerabilities.
- OpenAI is gating Astra’s most powerful cyber features to vetted Daybreak defenders, refusing proof‑of‑concept exploit requests in the public build, and plans broader paid ChatGPT and API rollouts in the coming days.
- The launch follows a summer of containment failures that prompted pauses and new safety checks, and it raises urgent policy and security questions because the same abilities that help defenders could be misused if external verification and stronger monitorability are not established.