Breaking

OpenAI Releases GPT-6 Astra as It Contests Anthropic’s Enterprise Lead

September 3, 2026 at 14:26 EDT

Foundation Models
AI Agents
Security

OpenAI has released GPT-6 Astra, a model the company presents as its most intelligent yet and as a generational leap for autonomous professional work. The launch also makes Astra the first OpenAI model designated as meeting the company’s Critical cybersecurity capability threshold, putting safety controls at the center of its rollout.

September 3, 2026 · OpenAI

OpenAI releases GPT-6 Astra

The new model is designed to work directly inside software, while its most advanced cybersecurity capabilities roll out under tighter controls.

The release in four numbers

100%

ExploitBench score

Zero-days found in testing

91.5%

Cyber jailbreak refusals

50%

Fewer manual fixes reported by Playco

Cyber results shown reflect Daybreak Blue access rather than the default production configuration.

Playco's reported manual-fix comparison

Relative volume: previous model = 100

100

Previous model

GPT-6 Astra

Takeaway: Playco reports half the manual fixes during game prototyping.

From goal to working software

Step 1

Set a goal

→

Step 2

Edit in software

→

Step 3

Test changes

→

Step 4

Refine output

What looks promising

Early use reports stronger spatial reasoning, improved interface responsiveness and more usable first-pass prototypes.

What remains constrained

Advanced cybersecurity access starts with testers, while monitoring may slow, pause or stop legitimate tasks.

Continue reading

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