Short Answer: What is GPT-6 Astra?
GPT-6 Astra is OpenAI’s newest flagship model, released September 3, which the company calls its “most intelligent and aligned model” yet. It posts state-of-the-art scores on math, coding, and computer-use benchmarks, but it is rolling out slowly — starting with a small group of cybersecurity-vetted organizations — because OpenAI says it crosses the “Critical” threshold for cyber capability under the company’s own safety framework.
Quick Summary
- Astra saturates FrontierMath Tier 4 (98%) and ARC-AGI-3 (99.9%), and scores 100% on ExploitBench, OpenAI’s own cyber-exploit benchmark.
- It sets a new bar on computer and browser use — controlling a Mac or desktop in the background to complete multistep work like spreadsheets and presentations.
- In Codex, Astra can keep persistent notes across long sessions instead of repeatedly summarizing and losing detail — rolling out as opt-in now, default “in the coming weeks.”
- Rollout is deliberately staged: cybersecurity-vetted organizations first, then ChatGPT Plus, Pro, Business, and Enterprise, plus the OpenAI API and AWS.
- This is the model that was previously paused for additional cyber-safety evaluation before this release.
9to5Mac first reported the details from OpenAI’s official announcement.
What exactly did OpenAI release?
GPT-6 Astra is OpenAI’s successor to the GPT-5 line, which launched in August 2025 and went through half a dozen point releases — 5.1, 5.2, 5.3, 5.4, 5.5, and July’s 5.6 Sol, Terra, and Luna — before this jump to a new major version. OpenAI describes Astra as “the world’s most intelligent and aligned model,” built on what it calls years of research across pre-training, reinforcement learning, and alignment.
The benchmark numbers are unusually strong even by frontier-model standards. Astra saturates FrontierMath Tier 4 with a 98% score, and OpenAI says it has already helped solve previously open problems in mathematics. It also saturates ARC-AGI-3, a benchmark specifically designed to resist memorization, at 99.9%, and hits a perfect 100% on ExploitBench, OpenAI’s internal benchmark for cybersecurity exploit tasks.
OpenAI is also positioning Astra as its strongest model yet for software engineering, and for “computer use” — letting ChatGPT operate a Mac or other desktop, including in the background while someone works, to handle multistep professional tasks like building spreadsheets and polished documents rather than just answering questions in a chat window.
What changed for developers using Codex?
Long coding sessions have historically forced models to periodically compress, or “compact,” their working context into a summary once the context window fills up — a process that can quietly drop details about why a fix failed or how a specific component behaves. Astra introduces an alternative: it can keep running notes across context windows without collapsing them into a single summary, and it can still search back through earlier context to find requirements or test results even if they never made it into its notes.
The feature is available now as an experimental, opt-in setting in Codex, and OpenAI says it will become the default behavior for Astra within Codex in the coming weeks.
Why is this rolling out so slowly?
OpenAI calls Astra “a significant jump in cyber capabilities” and says it meets the Critical threshold for cybersecurity risk under the company’s own Preparedness Framework — its highest defined risk tier. That is precisely why access is staged rather than immediate: Astra is going out today only to a limited set of organizations, with wider availability to ChatGPT Plus, Pro, Business, and Enterprise users, plus the OpenAI API and AWS, arriving “over the coming days.”
This is also, notably, the same model that OpenAI reportedly paused for additional cybersecurity evaluation before this release — so the cautious rollout looks less like standard process and more like a direct response to what that evaluation found. Enterprise administrators will need to explicitly turn Astra on for their workspace; it is off by default at launch, and Pro, Business, and Enterprise users also get access to a separate GPT-6 Astra Pro variant.
Why does this matter for AEO and brand marketing?
The headline benchmarks matter less here than the computer-use claim. A model that can operate a browser and desktop autonomously to complete multistep work is a model that can research, compare, and potentially transact on a user’s behalf without that user ever manually visiting a website — extending the same shift Anthropic’s Fable 5.1 leaned into just two days ago with its own push into long-running, tool-using agents. As agentic browsing gets more capable across every major lab at roughly the same time, being the source an agent actually trusts and cites while researching on a user’s behalf becomes a more urgent, and more literal, version of the AEO problem than it was when the only thing at stake was a chat answer.
The other detail worth watching is the safety framing itself. OpenAI naming a specific, elevated cyber-risk tier and staging rollout around it is a more formal, public version of the same posture other labs have taken this year after their own incidents. Expect more releases to ship with this kind of explicit risk-tier language attached, rather than as a footnote.
Frequently Asked Questions
What is GPT-6 Astra?
GPT-6 Astra is OpenAI’s newest flagship AI model, released September 3, 2026, succeeding the GPT-5 line. OpenAI describes it as its most intelligent and aligned model to date, with state-of-the-art results on math, coding, and computer-use benchmarks.
Why is GPT-6 Astra rolling out slowly instead of to everyone at once?
OpenAI says Astra meets the “Critical” cybersecurity threshold under its Preparedness Framework, its highest defined risk category. Access is starting with a limited set of vetted organizations before expanding to ChatGPT Plus, Pro, Business, and Enterprise users over the following days.
Is GPT-6 Astra available to regular ChatGPT users?
Not immediately. It launched first to a limited group of organizations. OpenAI says it will become available to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS, over the following days.
What is new for Codex specifically?
Astra can retain notes across long coding sessions instead of repeatedly compressing context into summaries, which historically caused models to lose details about why a fix failed or how code behaves. It also remains able to search earlier context directly. The feature is opt-in now and becomes Codex’s default for Astra in the coming weeks.
How does GPT-6 Astra relate to GPT-5.6?
GPT-5.6, released as Sol, Terra, and Luna variants in July, was the last update in the GPT-5 line before this jump to GPT-6. See our coverage of the GPT-5.6 launch for how that release compared.
Source: 9to5Mac.
About the author
Kai Williams
Kai Williams has been in marketing for years, with a long background in SEO before AEO had a name. He stepped into Answer Engine Optimization the moment AI started reshaping how people search, and has been tracking the shift ever since. At Prompt Insider, he covers AEO, AI marketing, and the future of search, breaking down what is changing and what brands need to do about it.
