OpenAI has released GPT-6 Astra, a new flagship model designed for difficult, long-running work across reasoning, software engineering, computer use, research, and document creation.
The short answer: Astra is a major capability release, but it is not available to every ChatGPT user at the same moment. OpenAI says the rollout begins with enterprises in its Trusted Access Program. ChatGPT Plus, Pro, Business, and Enterprise customers—as well as API, Codex, and AWS customers—are expected to receive access in the coming days.
This guide explains what GPT-6 Astra is, what changed, how much the API costs, who gets access first, and why the model’s cybersecurity safeguards are a central part of the launch.
Last verified: September 3, 2026. Rollout details and model availability can change quickly, so confirm access in your own ChatGPT workspace or API project.
GPT-6 Astra at a glance
According to the official GPT-6 Astra model documentation, the model has:
- a 1,050,000-token context window;
- up to 128,000 output tokens;
- text input and output plus image input;
- reasoning-effort controls for
low,medium,high,xhigh, andmax; - support for the Responses API and Chat Completions API;
- function calling and structured outputs;
- built-in support for web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search;
- an April 30, 2026 knowledge cutoff.
OpenAI describes Astra as its most capable model for hard end-to-end work. The launch video emphasizes professional computer tasks, software engineering, and stronger alignment rather than positioning Astra as a simple chat-only upgrade.
Is GPT-6 Astra available in ChatGPT?
Yes—but access is rolling out in phases.
The official OpenAI launch video says GPT-6 Astra is initially rolling out to enterprises and customers in the Trusted Access Program. Broader availability in ChatGPT Work and Codex is planned for Plus, Pro, Business, and Enterprise plans, with API and AWS access following in the next few days.
That means two people on the same paid plan may not see Astra at exactly the same time. If it is not yet visible in your model selector, that does not necessarily indicate an account problem.
OpenAI also recommends using the ChatGPT desktop app to experience Astra’s computer-use capabilities at their best. The practical advice is to check the desktop app, confirm your workspace policy allows the model, and wait for the phased rollout rather than relying on unofficial workarounds.
What is new in GPT-6 Astra?
1. Long-running computer use
Astra is designed to complete multi-step work across professional and desktop applications. This is more ambitious than answering a prompt: the model can inspect context, use tools, take actions, check results, and continue toward an outcome.
For businesses, that can make Astra relevant to workflows such as:
- reviewing and organizing large document sets;
- navigating internal tools and operational dashboards;
- creating reports, presentations, and structured deliverables;
- performing multi-step research with citations;
- testing software through a browser or desktop interface;
- carrying out repetitive knowledge-work processes with human approval.
Computer use still needs clear permissions, restricted environments, audit logs, and review points. A stronger model does not remove the need for operational controls.
2. Stronger software engineering
OpenAI calls Astra its best software-engineering model to date. The model page supports that positioning with tools for hosted shell access, patch application, coding, computer use, and long-context work.
The most useful improvement is likely not isolated code completion. It is the ability to understand a larger codebase, plan a change, edit several connected files, run tests, inspect failures, and iterate. Teams should evaluate this on their own repositories because benchmark strength does not automatically equal production reliability.
3. A context window above one million tokens
The 1,050,000-token context window can hold substantially more source material in a single request than previous mainstream models. That creates room for larger repositories, long policy sets, extensive research material, and multi-document analysis.
More context is not automatically better. Very large prompts can increase latency, cost, and the amount of irrelevant material the model must process. OpenAI’s pricing page notes that prompts above 272,000 input tokens use higher rates, so retrieval and context selection remain important engineering decisions.
4. Broad tool support
Astra supports the tools needed to build capable AI agents: web and file search, code execution, hosted shell, patching, skills, computer use, MCP connections, and tool discovery.
This matters because the value of an agent depends on more than model intelligence. It needs dependable tools, narrow permissions, error handling, evaluation, monitoring, and a safe way to ask for human approval when an action has consequences.
GPT-6 Astra API pricing
OpenAI’s published standard text-token pricing is:
| Token type | Price per 1 million tokens |
|---|---|
| Input | $10.00 |
| Cached input | $1.00 |
| Cache writes | $12.50 |
| Output | $50.00 |
Prompts with more than 272,000 input tokens are billed at twice the input and cache rates and 1.5 times the output rate for the full request. Batch and Flex processing are listed at 50% of Standard rates, while Fast mode is priced at twice the applicable rate.
For production planning, calculate cost per completed workflow—not only cost per token. A more expensive model can still be economical if it completes a difficult task with fewer retries, less manual correction, or less orchestration. The reverse is also true: using Astra for simple classification or templated extraction may be unnecessary.
Why safety is a major part of the Astra release
The GPT-6 Astra System Card says Astra is the first OpenAI model to reach the Critical level for cybersecurity capability under the company’s Preparedness Framework. OpenAI says it strengthened safeguards against harmful cyber actions and added tighter controls around internal development and deployment.
The reported measures include defense-in-depth safeguards, updated safety training for agentic cyber scenarios, expanded monitoring, more conservative behavior boundaries for higher-risk accounts, and phased trusted access for advanced defensive work.
This does not mean normal users should avoid the model. It means organizations should take deployment governance seriously—especially when Astra can access code, terminals, cloud systems, internal files, or production applications.
A practical deployment should include:
- least-privilege accounts and scoped credentials;
- isolated test or sandbox environments;
- approval gates for external or destructive actions;
- complete action logs and alerting;
- representative safety and quality evaluations;
- clear escalation paths when the model is uncertain;
- human review for legal, financial, security, and other high-impact decisions.
Should your business use GPT-6 Astra?
Consider Astra when a workflow is genuinely complex: large context, many connected steps, difficult reasoning, software engineering, tool use, or cross-application computer work. It may be a strong fit for agentic development, technical research, document-heavy professional services, internal operations, and advanced defensive security work.
Start with a bounded evaluation rather than moving every workflow immediately. Select 20 to 50 representative tasks, define what a correct result looks like, record latency and total token cost, and compare Astra with the model you use today. Include failure cases and the cost of human correction.
For simpler or high-volume tasks, a smaller model may remain the better choice. Good AI architecture routes each task to the least expensive model that meets the required quality and risk threshold.
A practical GPT-6 Astra adoption checklist
- Confirm access. Check the model selector or API project instead of assuming rollout availability.
- Choose one valuable workflow. Avoid a company-wide migration before you have evidence.
- Build a baseline. Record the current model’s accuracy, completion rate, latency, and cost.
- Test representative tasks. Include difficult examples and known failure cases.
- Limit permissions. Give the agent only the files, systems, and actions it needs.
- Add human approval. Require review before external messages, production changes, purchases, deletions, or sensitive decisions.
- Monitor full outcomes. A tool call succeeding does not prove the business task was completed correctly.
- Review cost at scale. Include long-context pricing, retries, tool fees, and human correction.
What GPT-6 Astra means for AI teams
GPT-6 Astra moves the centre of gravity from short conversations toward longer, end-to-end work. Its million-token context window, software-engineering focus, computer use, and broad tool support make it especially relevant to teams building AI agents and operational systems.
The opportunity is significant, but the implementation still determines the result. Model capability needs to be paired with careful context design, reliable integrations, evaluations, permissions, monitoring, and human oversight.
If you want to evaluate GPT-6 Astra for a real workflow, I provide AI consulting and custom AI development covering use-case selection, agent architecture, OpenAI API integration, safety controls, evaluation, and production deployment. You can also book a free strategy call to discuss the best first pilot for your team.
Official sources
Frequently asked questions
What is GPT-6 Astra?
GPT-6 Astra is OpenAI's most capable model for difficult end-to-end work, including complex reasoning, coding, computer use, research, and document creation. It supports text input and output, image input, tool use, and multiple reasoning-effort levels.
Is GPT-6 Astra available in ChatGPT now?
OpenAI says Astra is rolling out first to enterprises in its Trusted Access Program. Access through ChatGPT Plus, Pro, Business, and Enterprise plans is expected in the coming days, so availability may differ by account and region during the phased rollout.
Is GPT-6 Astra available through the OpenAI API?
OpenAI says API access is coming in the next few days. The documented model ID is gpt-6-astra, and the model supports both the Responses API and Chat Completions API. Developers should confirm access in their own project before changing production traffic.
How much does the GPT-6 Astra API cost?
The published standard text-token rates are $10 per million input tokens, $1 per million cached input tokens, $12.50 per million cache-write tokens, and $50 per million output tokens. Longer prompts and faster processing modes can change the effective price.
What is GPT-6 Astra's context window?
GPT-6 Astra has a 1,050,000-token context window and supports up to 128,000 output tokens. Prompts above 272,000 input tokens use higher token pricing, so teams should test quality, latency, and cost with representative workloads.