OpenAI launched GPT-6 Astra on September 3, 2026, with a 1M-token context window, computer use, and the industry's first Critical-level cyber safeguards. Here is what changed and what it costs.
By Quantara Editorial Team · AI Tools · Published · 9 min read
OpenAI announced GPT-6 Astra on September 3, 2026, describing it as "the most intelligent and aligned model in the world." A limited preview shipped to trusted partners first, with a broader rollout to ChatGPT Plus, Pro, Business, and Enterprise, plus the OpenAI API and AWS, following over the next several days. The launch also arrived under an unusual amount of scrutiny: Astra is the first OpenAI model to cross what the company calls the Critical threshold for cybersecurity capability under its Preparedness Framework.
GPT-6 Astra: Quick Facts
| Detail | GPT-6 Astra |
|---|---|
| API model ID | gpt-6-astra |
| Context window | 1,000,000 tokens |
| Standout capability | Computer use — navigating a computer interface like a human |
| API price per 1M tokens | $10 input / $50 output (cached input $1) |
| Price vs. previous flagship | Roughly 2.5x GPT-5.6 Sol's pricing |
| Safety classification | First OpenAI model at the Critical cybersecurity threshold |
| Rollout | Staged: trusted partners first, then ChatGPT tiers and the API |
What Astra Can Do
OpenAI positions Astra as state-of-the-art across computer use, browsing, software engineering, cybersecurity, science, and professional work. The headline feature is computer use: Astra can operate a desktop environment directly, clicking, typing, and navigating applications the way a person would, rather than only calling predefined tools or APIs. OpenAI president Greg Brockman framed the release as a meaningful step toward more autonomous, general-purpose agents.
Pricing and Access
Astra costs $10 per million input tokens and $50 per million output tokens on the API, with cached input priced at $1 per million tokens. Batch and flex processing are available at half price for latency-tolerant workloads, while a Fast mode doubles the standard rate for lower-latency responses. That pricing is roughly 2.5 times GPT-5.6 Sol's rate card, reflecting both the jump in capability and the additional safety infrastructure OpenAI built around this release.
Access is staged rather than immediate for everyone. A limited set of organizations received Astra on day one, with ChatGPT Plus, Pro, Business, and Enterprise customers, plus the OpenAI API and AWS, gaining access over the following days.
Why the Rollout Was Delayed
OpenAI had already pushed back its next flagship release once, after a data-exposure incident involving a Hugging Face-hosted resource in July 2026 prompted the company to add more safeguards before shipping. Astra's launch carried a second layer of caution: internal evaluations found the model could identify and develop functional exploits against hardened real-world systems without human help, and had discovered two previously unknown zero-day vulnerabilities during testing. That combination is what triggered OpenAI's Critical threshold under its own Preparedness Framework, and it meant additional safeguards had to be built into training itself, not bolted on right before release.
The commercial version of Astra ships with restrictions on its most advanced offensive security capabilities, and OpenAI says certain high-risk requests are blocked outright rather than answered.
What This Means for Teams Using Astra
- Expect higher per-token costs than GPT-5.6 Sol, offset by stronger reasoning, browsing, and computer-use capability for complex tasks.
- Computer-use workflows (form filling, multi-app research, QA testing) are the most differentiated new use case — evaluate those before assuming Astra is simply "a smarter chat model."
- Security teams should read OpenAI's safety overview before enabling Astra for anything touching production infrastructure, given its Critical cyber classification.
- Rollout is staged, so API and Enterprise access may lag the initial announcement by several days.
The Bottom Line
GPT-6 Astra is OpenAI's most capable model to date, built around a genuinely new capability — computer use — rather than just incremental benchmark gains. It is also the company's first model serious enough, from a cybersecurity standpoint, to require a new tier of internal safeguards before shipping at all. For most teams, that means real capability gains, a real price increase, and a real reason to read the safety documentation before turning on computer-use features in production.
Sources
- OpenAI: Path to Astra — critical capabilities and frontier safeguards
- OpenAI: Safety overview for GPT-6 Astra
- CNBC: OpenAI begins rolling out Astra model after warning of its advanced cyber capabilities
- Fortune: OpenAI launches GPT-6 Astra, touts computer use
- 9to5Mac: OpenAI releasing major upgrade to ChatGPT and Codex with GPT-6 Astra
Frequently asked questions
- When did GPT-6 Astra launch?
- OpenAI announced GPT-6 Astra on September 3, 2026, shipping first to a limited set of trusted partners before rolling out to ChatGPT Plus, Pro, Business, and Enterprise, plus the API and AWS.
- How much does GPT-6 Astra cost?
- Astra costs $10 per million input tokens and $50 per million output tokens on the API, with cached input at $1 per million tokens. That is roughly 2.5x the price of GPT-5.6 Sol.
- What is Astra's "Critical" cybersecurity rating?
- Under OpenAI's Preparedness Framework, a model reaches the Critical threshold if it can independently develop functional exploits against hardened real-world systems. Astra is the first OpenAI model to cross that line, which is why its release required extra safeguards built in during training.
- What is computer use in GPT-6 Astra?
- Computer use lets Astra operate a desktop environment directly — clicking, typing, and navigating software the way a person would — instead of relying only on predefined tool calls or APIs.
- Can I use Verbito to write prompts for GPT-6 Astra?
- Yes. Verbito generates structured prompts you can adapt for Astra, including role, context, task, and output-format sections that work well with computer-use and agentic workflows.