Breaking

OpenAI Releases GPT-6 Astra With Advanced Computer Use and Critical Cybersecurity Capability

September 4, 2026 at 01:38 EDT

Foundation Models
AI Agents
Security

OpenAI released GPT-6 Astra on Sept. 3, introducing a flagship model built for computer use, coding, browsing, scientific work and multi-step professional tasks. The launch also makes Astra the first OpenAI model designated at the “Critical” cybersecurity capability level under its Preparedness Framework, so its most powerful cyber functions will be staged and gated. OpenAI’s launch overview

SEP 3, 2026 · OPENAI

GPT-6 Astra brings computer use to center stage

OpenAI’s new flagship model combines reasoning, coding, browsing and professional workflows. Its strongest cybersecurity capability has also crossed the Critical threshold, bringing tighter access and monitoring.

At a glance

72.6%

OSWorld 2.0 score

97.6%

FrontierMath Tier 4

100.0%

ExploitBench score

1,050,000

Token context window

OSWorld 2.0 · task completion

Higher columns indicate a higher score on the same benchmark.

GPT-5.6 Sol

65.7%

GPT-6 Astra

72.6%

Astra scored higher and completed the simulated tasks in about 47% less time.

THE DEPLOYMENT TRADE-OFF

More agency also means more guardrails.

Practical upside

Strong backend engineering, computer-use workflows and clearer progress updates.

Operational limit

Advanced exploit work is gated; safety checks can pause legitimate tasks, and monitorability remains an open issue.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.