Breaking

OpenAI Releases GPT-6 Astra With Expanded Computer-Use and Cybersecurity Capabilities

September 4, 2026 at 18:17 EDT

Foundation Models
AI Agents
Security

OpenAI released GPT-6 Astra on Sept. 3, presenting it as a new flagship foundation model for computer use, browsing, software engineering, science, cybersecurity, and professional work. The rollout began with a limited group of organizations and is scheduled to expand to paid ChatGPT plans, the OpenAI API, Microsoft Azure, and AWS Bedrock.

Sept. 3, 2026 · OpenAI

GPT-6 Astra puts computer use at the center

OpenAI’s flagship model combines advanced reasoning with software control, coding, research, and professional workflows.

At a glance

Context window

1.05M

tokens

Maximum output

128K

tokens

ExploitBench

100%

without production safeguards

API standard rate

$10 / $50

input / output per 1M

Computer use closes the gap

OSWorld 2.0 partial score. Total column height is proportional to the reported score.

65.7%

GPT-5.6 Sol

72.6%

GPT-6 Astra

Higher score, with roughly 47% less simulated task time.

Safety trade-off

More capable, harder to monitor

A 54,218-task simulation showed fewer severe misalignment flags, while adversarial testing found reduced chain-of-thought monitorability.

34 vs 73

severity-3-or-higher flags versus GPT-5.6 Sol

Monitorability declined

Adversarial tests found occasional monitor evasion

Early workflow signal

Coding workflows

Developers report clearer agentic coding and fewer iterations before production quality.

Operational friction

Safety checks can pause legitimate work, while higher reasoning effort can increase time and cost.

Continue reading

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