On September 3, 2026, OpenAI began rolling out GPT-6 Astra, a new flagship artificial intelligence model that the company is calling its most capable release to date. Unlike previous updates that mostly improved conversational ability, Astra has been built around a different goal: letting an AI system operate a computer the way a human would — clicking buttons, filling out forms, navigating browsers, and completing multi-step professional tasks with minimal supervision. For anyone following the artificial intelligence space, GPT-6 Astra marks one of the most significant product launches of the year, and it raises fresh questions about how quickly AI agents will move from novelty tools into everyday business infrastructure.
In this article, we break down what GPT-6 Astra actually is, how it differs from earlier models, what it can do, how much it costs, and what the rollout means for developers, businesses, and everyday users of ChatGPT.
What Is GPT-6 Astra?
GPT-6 Astra is OpenAI’s newest large language model, positioned as a generational leap over GPT-5.6 “Sol” and earlier GPT-5-series models. OpenAI President Greg Brockman described the release as a turning point for the company, suggesting that when the history of artificial general intelligence is eventually written, this moment may be remembered as a milestone on that path. That kind of language is not unusual from AI labs eager to promote their newest release, but the underlying capabilities behind Astra do represent a meaningful jump from what ChatGPT users have grown used to.
At a technical level, Astra supports a context window of roughly 1.05 million tokens and can generate outputs of up to around 128,000 tokens in a single response. That means the model can hold an enormous amount of information — entire codebases, lengthy legal contracts, or hours of meeting transcripts — in working memory while producing detailed, long-form output. Astra also supports adjustable “reasoning effort,” letting developers dial the model’s thinking time up or down depending on whether they need a fast answer or a deeply considered one.
Computer Use and Agentic Capability
The headline feature of GPT-6 Astra is its ability to use computers autonomously. OpenAI says the model can fill out online forms, update customer relationship management records, organize calendars, conduct research, draft summaries, analyze scientific data, generate plots, and build and test websites. According to reporting from The Guardian, OpenAI has claimed the model can complete tasks such as a job search in under three minutes that would otherwise take a human several hours.
This is a significant shift in how AI models are being marketed. Rather than positioning Astra purely as a chatbot or writing assistant, OpenAI is pitching it as a digital worker capable of operating software interfaces directly, rather than relying solely on APIs. In practice, this means Astra can look at a screen, understand what it sees, and interact with it much like a person using a mouse and keyboard would. On a benchmark called OSWorld 2.0, which measures how well an AI model can operate real desktop software, OpenAI reported a score of 72.6% for Astra, compared with 65.7% for the previous GPT-5.6 Sol model.
Astra is also being pitched as OpenAI’s strongest model yet for producing documents, spreadsheets, and slide decks that match a user’s existing writing and visual style, rather than generating generic templates. A new “Sites” feature inside ChatGPT allows users to create, host, and share websites, web apps, and simple games directly from a text prompt, without needing to write code themselves.
Benchmark Performance
OpenAI has published a series of benchmark results to support its claim that Astra is its most capable model so far. The company says Astra scores 98% on FrontierMath Tier 4, a benchmark designed to test advanced mathematical reasoning, and claims the model has already contributed to solving previously unsolved mathematical problems. On ARC-AGI-3, a benchmark often used to test general reasoning and novel problem-solving, OpenAI reports a score of 99.9%. On ExploitBench, a benchmark related to cybersecurity and vulnerability analysis, the company says Astra achieved a perfect 100% score.
These figures should, as always, be read with some caution — benchmark scores reported directly by the company that built the model are not independently verified in the same way third-party evaluations are. Still, the scale of the jump being claimed across math, reasoning, and security-related benchmarks suggests OpenAI has made substantial investments in both training data and computing infrastructure for this release.
Rollout Timeline and Availability
GPT-6 Astra’s rollout has been staged rather than immediate. On launch day, access was limited to a small group of organizations under OpenAI’s Trusted Access Program, along with a set of vetted cybersecurity professionals granted access through the company’s Daybreak initiative — a program specifically designed to let security researchers use frontier AI models for tasks like malware analysis and threat detection before the broader public gets access.
In the days following the initial launch, OpenAI has said it will expand access to ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as through the OpenAI API. Microsoft has also begun offering the model through its Foundry limited-access program on Azure, and Amazon has made Astra available through Amazon Bedrock and AWS for enterprise customers. Notably, OpenAI has not announced a timeline for bringing Astra to its Free or Go subscription tiers, meaning casual users may need to wait or upgrade to a paid plan to try the new model.
Pricing
For developers building on top of the model through OpenAI’s API, GPT-6 Astra pricing works out to roughly $10 per million input tokens and $50 per million output tokens under the standard pricing tier. Cached input tokens are considerably cheaper, priced at around $1 per million tokens, while writing new data into the cache costs about $12.50 per million tokens. OpenAI also offers a “Flex” and batch processing mode priced at roughly half the standard rate for less time-sensitive workloads, and a faster “Fast” mode priced at roughly double the standard rate for applications that need near-instant responses.
This tiered pricing structure reflects a broader trend across the AI industry: as models become more capable and more expensive to run, providers are increasingly offering multiple service levels so that businesses can balance cost against speed and quality depending on the specific task.
Why Astra Matters for Cybersecurity
One detail from the Astra rollout worth highlighting is OpenAI’s decision to grant early access to cybersecurity professionals through its Daybreak program before opening the model more broadly. This reflects a growing recognition across the AI industry that highly capable “computer use” models carry dual-use risk — the same abilities that let an AI model fill out a tax form or debug software could, in principle, be misused to probe for vulnerabilities or automate attacks. By giving trusted defenders early access to build detection tools and understand the model’s capabilities, OpenAI appears to be trying to get ahead of potential misuse before Astra becomes widely available.
OpenAI has also said Astra is its most reliable model to date in terms of understanding user intent, exercising caution on consequential decisions, and communicating clearly when it lacks the information needed to complete a task. As AI models are increasingly used to automate real business workflows rather than simply answer questions, this kind of built-in caution becomes an important safety consideration, not just a nice-to-have feature.
What This Means for the Broader AI Race
GPT-6 Astra’s release lands at a moment of intense competition among frontier AI labs, with OpenAI, Anthropic, and Google all pushing hard into agentic AI — systems designed to complete multi-step tasks autonomously rather than simply respond to single prompts. Some industry commentary ahead of the launch suggested OpenAI had been reallocating significant computing resources toward this release specifically to keep pace with rivals in pre-training scale. Whatever the internal motivations, the practical effect is that the gap between “AI chatbot” and “AI coworker” appears to be narrowing quickly, with each major lab racing to demonstrate that its model can handle real-world professional tasks, not just answer trivia or write essays.
For businesses evaluating whether to adopt GPT-6 Astra, the practical question is less about benchmark scores and more about reliability in specific workflows: how consistently the model completes multi-step tasks correctly, how well it handles edge cases and errors, and how much human oversight is still required. Early access through Business and Enterprise tiers should give organizations a chance to test these questions directly over the coming weeks.
Frequently Asked Questions
When was GPT-6 Astra released? OpenAI began rolling out GPT-6 Astra on September 3, 2026, starting with a limited group of trusted organizations before expanding to paid ChatGPT tiers and API access.
How much does GPT-6 Astra cost? Through the OpenAI API, standard pricing is approximately $10 per million input tokens and $50 per million output tokens, with cheaper rates available for cached input and batch processing.
Is GPT-6 Astra available for free ChatGPT users? As of the initial rollout, OpenAI had not announced availability for Free or Go tier ChatGPT users. Access has so far been limited to Plus, Pro, Business, and Enterprise subscribers, along with API and cloud partners.
What makes Astra different from previous GPT models? Astra’s main distinguishing feature is its ability to autonomously operate software interfaces — navigating websites, filling forms, and completing multi-step professional workflows — rather than only generating text responses.
Conclusion
GPT-6 Astra represents OpenAI’s most ambitious attempt yet to move AI models beyond conversation and into genuine task execution. With strong benchmark claims across math, reasoning, and computer-use tasks, a staged rollout designed to manage security risk, and expanding availability across ChatGPT, the API, Azure, and AWS, Astra is likely to shape how businesses think about AI automation over the coming months. As with any major model release, the real test will come from how it performs in everyday use — and how quickly rival labs respond with their own next-generation systems.
