OpenAI Releases GPT-6 Astra, First Model to Cross Its 'Critical' Cyber Threshold

OpenAI announced GPT-6 Astra on September 3, 2026, rolling it out first to a limited set of organizations. The company frames the model less as a chatbot than as a computer-use system: Astra operates browsers, spreadsheets, desktop applications and terminals the way a person does, completing multistep jobs rather than describing them. Access will expand over the coming days to ChatGPT Plus, Pro, Business and Enterprise users, and through the OpenAI API and AWS. For enterprise workspaces, access is off by default at launch.

Published evaluation results put numbers behind the jump. Astra scored 97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3 and 96.0% on GPQA Diamond. On ARC-AGI-3, its predecessor GPT-5.6 Sol managed only 7.8%. On OSWorld 2.0, which measures computer use, Astra reached 72.6% in roughly 40 minutes per task, compared with 65.7% in roughly 75 minutes for Sol. It posted 57.9% on Terminal-Bench 4.0 and 92.7% on ScreenSpot-Pro. In the API the model is served as gpt-6-astra at $10 per million input tokens and $50 per million output tokens — 2.5 times the price of GPT-5.6 Sol.

The cybersecurity findings are the most contentious part of the release. OpenAI designated Astra the first model to meet the "Critical" cybersecurity threshold under its Preparedness Framework. Tested without production safeguards, it scored 100% on ExploitBench and 42.4% on ExploitGym, against 78.5% and 30.3% for GPT-5.6 Sol. To rule out benchmark contamination, the company built a fresh evaluation from vulnerabilities disclosed over the previous three months, where Astra reached 39.0%. During that evaluation the model discovered and used two previously unknown zero-day vulnerabilities, which OpenAI said it disclosed to the affected maintainers. Advanced offensive cyber tasks are therefore blocked in the shipping version, with wider access planned through the company's Daybreak program.

OpenAI president Greg Brockman called the model a generational leap and said it could eventually be seen as the arrival of artificial general intelligence. Independent evaluators read the picture more cautiously. Artificial Analysis places Astra at 61 on its Intelligence Index, level with GPT-5.6 Sol and five points behind Claude Fable 5.1, while its Coding Agent Index score of 67 matches Fable 5 at less than half the cost per task. The same analysis records a drop in hallucination rate from 92% to 51%, alongside regressions of two to three points on several other evaluations. OpenAI itself acknowledged that Astra's written reasoning proved harder to monitor than its predecessor's, and described improving monitorability as an ongoing research priority.