OpenAI Releases GPT-6 Astra: The World's Most Capable Model with Critical Cybersecurity

OpenAI has released GPT-6 Astra, the company's most capable model ever broadly deployed. The announcement comes days after OpenAI disclosed that Astra meets the Critical cybersecurity capability threshold under its Preparedness Framework.

What's New

GPT-6 Astra represents a significant leap forward in several key areas:

Cybersecurity

• First Critical-tier model: Astra is OpenAI's first model designated at the Critical cybersecurity capability level
• ExploitBench: Achieved 100% success rate (vs 78.5% for GPT-5.6 Sol)
• ExploitGym: 42.4% success rate (vs 30.3% for GPT-5.6 Sol)
• SRE-Bench: 88.0% single-attempt success (vs 55.9% for GPT-5.6 Sol)
• Capable of identifying and developing zero-day exploits across many well-protected systems

Computer Use & Coding

• AutomationBench: 41.4% (vs 18.1% for GPT-5.6 Sol)
• BenchCAD: 95.9% (vs 83.3% for GPT-5.6 Sol)
• BrowseComp: 91.5%
• Can create documents, spreadsheets, and presentations from templates
• Adapt to new requirements mid-task

Long Context

• Supports extended context windows for complex, multi-step reasoning
• Performs well on long-form document analysis

Pricing

• Input: $10 per million tokens
• Output: $50 per million tokens
• Cached input: $1 per million tokens
• Cache writes: $12.50 per million tokens

Availability

GPT-6 Astra is rolling out today to a limited set of organizations, with broader availability planned over the coming days to:

• ChatGPT Plus, Pro, Business, and Enterprise users
• OpenAI API customers
• Microsoft Azure customers
• AWS Bedrock customers

Safety & Safeguards

Given Astra's Critical cybersecurity capabilities, OpenAI has implemented significantly strengthened protections:

• Stronger model robustness to jailbreaks
• Enhanced context for monitoring systems
• Stricter isolation for development and deployment
• Checkpoint encryption
• Universal monitoring of full trajectories including chains of thought
• Blocking alignment evaluation process before internal use

Why This Matters for Builders

For security teams: Astra's capability means defenders can find vulnerabilities faster, but the Critical designation signals that these capabilities require careful governance. The Daybreak program provides a pathway for defensive use.

For developers: Astra's coding and computer-use performance represents a major step forward for autonomous agent workflows, particularly for complex, multi-step tasks.

For enterprises: The rollout to ChatGPT Pro/Business/Enterprise plans means teams can access these capabilities without new subscriptions, though usage is included in allowances.

What's Not Here

This is a model release announcement. The underlying infrastructure (ChatGPT, Codex, API) remains unchanged. The most advanced cybersecurity workflows stay limited initially through the Daybreak program.

• https://openai.com/index/gpt-6-astra/
• https://openai.com/index/safety-overview-gpt-6-astra/
• https://openai.com/index/path-to-astra/