OpenAI Ships GPT-6 Astra With 1M-Token Context and Major Cybersecurity Gains

OpenAI announced the release of GPT-6 Astra on September 3, 2026, describing it as 'the most intelligent and aligned model in the world.' The model carries the API identifier gpt-6-astra and ships with a 1-million-token context window.

What's New

• Major improvements in coding, research, and complex multi-step work
• Computer-use tasks run nearly 2x faster than the prior model generation
• Significant gains specifically in cybersecurity resilience, jailbreak resistance, browsing safety and handling of higher-risk scenarios
• 1-million-token context window, up substantially from prior models

Pricing

On the API, GPT-6 Astra is priced at $10 per million input tokens and $50 per million output tokens — reported as roughly 2.5x the pricing of OpenAI's prior GPT-5.6 Sol model. Cached input is priced at $1 per million tokens, batch and flex processing are available at half the standard rate, and a Fast mode is available at double the standard rate.

The Cyber-Risk Delay Context

Notably, this release comes roughly four weeks after OpenAI disclosed that it had slowed the model's development and rollout specifically over cyber-risk concerns. The heavy emphasis on cybersecurity resilience, jailbreak resistance and browsing safety in this release's stated improvements directly addresses that earlier disclosed concern, rather than being presented as a routine capability upgrade alone.

Rollout Timeline

Rollout is staged rather than immediate and universal: a limited set of organizations gain access first, followed by ChatGPT Plus, Pro, Business and Enterprise plan holders over the following days, alongside availability through the OpenAI API and AWS.

What This Means for Enterprise AI Adopters

For organizations already building on OpenAI's API, the staged rollout and 2.5x price increase over the prior model are the two practical planning items — budget and access timing should be evaluated before assuming immediate, blanket availability. The specific emphasis on cybersecurity and jailbreak resistance, tied directly to a previously disclosed delay, is also a useful data point for organizations evaluating AI vendor safety practices more broadly, not just this one release.