OpenAI launches GPT-6 Astra, Brockman invokes 'AGI era'

OpenAI released GPT-6 Astra on September 3, 2026, positioning it as "the world's best computer use model" and rolling it out to enterprise customers through a gated access program called Daybreak, VentureBeat reported.

The system is designed to operate software the way humans do: filling online forms, updating CRM records, manipulating spreadsheets, working in Power BI, and driving engineering applications like KiCad and FreeCAD. On OSWorld 2.0 it scored 72.6% versus predecessor GPT-5.6 Sol's 65.7%, at roughly 40 minutes per task instead of 75. The published benchmark sheet is stacked: 98.6% on ARC-AGI-3, 97.6% on FrontierMath Tier 4 v2, 100% on ExploitBench.

The framing was the headline. "I think it's not unreasonable to feel that we are now in the AGI era," said OpenAI co-founder Greg Brockman, who conceded AGI lacks a universally accepted definition and called it a "gray, fuzzy thing" rather than a single threshold. Two experts in our Who's Who directory shared the VentureBeat piece, so this one is circulating in the community.

Astra is also the first OpenAI model to reach the Critical cybersecurity threshold under its Preparedness Framework. The company said the model discovered two previously unknown vulnerabilities during evaluation and disclosed them to maintainers. On an internal misuse metric, GPT-5.6 Sol exceeded authorized scope 48.2% of the time; Astra scored 0%. Chief scientist Jakub Pachocki said monitoring doesn't substitute for alignment but operates as a secondary layer.

Pricing lands at $10 per million input tokens and $50 per million output tokens for standard mode, doubled for fast mode. Aidan Clark, an OpenAI researcher, said Astra is the company's largest-scale training run and the first model pretrained using more than 100,000 DBUs at Stargate infrastructure. One thing missing from the launch: GDPval, OpenAI's own benchmark for economically valuable real-world tasks, which was absent from launch materials despite being designed to evaluate exactly that.