GPT-6 Astra: Who Can Use It, Price and What Is Rumour
- 15 views
- 0 likes
- 0 comments
- All languages combined

In short
OpenAI announced GPT-6 Astra on 3 September 2026, first for a limited set of organizations and then for paid ChatGPT plans and the API at $10 and $50 per million tokens [1][7]. Plus users get it in Work and Codex; Free users do not [2]. GPT-6.2 Astra is an unconfirmed rumour [9].
Key takeaways
- GPT-6 Astra was announced on 3 September 2026 and is OpenAI's most capable model [1].
- It started with a limited set of organizations, Daybreak companies per CNBC, then expanded to paid ChatGPT plans, the API, Azure and Bedrock [1][7].
- Pro, Business and Enterprise get GPT-6 Pro powered by Astra in Chat; Plus gets Astra only in Work and Codex; Free and Go get none [2].
- API price: $10 per million input and $50 per million output tokens [1].
- Astra is OpenAI's first model rated Critical for cyber risk; reduced-refusal cyber use is for Daybreak Red organizations only [1][3].
- "Welcome to the AGI era" was Greg Brockman's view, and GPT-6.2 Astra is only a rumour [8][9].
On this page 11 sections
What is GPT-6 Astra?
GPT-6 Astra is OpenAI's most capable model, announced on 3 September 2026 [1][7]. OpenAI calls it "the world's most intelligent and aligned model" and says it leads on computer use, browsing, software engineering, cybersecurity, science and professional work [1]. Those are OpenAI's own claims, based mostly on its own choice of tests.
Astra is the top of a larger family. Since September OpenAI has added GPT-6 Sol and GPT-6 Luna (22 September), GPT-6.1 Sol (29 September) and GPT-6.1 Terra (26 October) [1][4][5]. Before that, on 9 July 2026, OpenAI had launched the GPT-5.6 family, Sol, Terra and Luna, together with ChatGPT Work, an agent mode for multi-step work across apps and files [6]. Those July products shipped and are still offered [2].
Was Astra limited to Daybreak?
Only at the very start. OpenAI said on launch day that Astra was rolling out first to a limited set of organizations and, over the following days, would become available to all ChatGPT Plus, Pro, Business and Enterprise users, and through the OpenAI API, Microsoft Azure and AWS Bedrock [1]. CNBC reported that the first group was companies in Daybreak, OpenAI's application-based cybersecurity programme [7].
What stays restricted is Astra's most sensitive cyber use. In standard use, OpenAI says Astra refuses advanced security tasks such as building proof-of-concept exploits [1]. Reduced refusals on Astra are available only in the Daybreak Red tier, which OpenAI currently offers only to approved business and enterprise organizations, not to individuals [3]. The lower Daybreak Blue tier keeps standard safeguards on Astra [3].
So the accurate summary is: Astra is generally available to paying customers, while its unrestricted cyber capabilities sit behind Daybreak Red.
Who can use GPT-6 Astra now?
Access depends on the plan and on which part of ChatGPT you use. According to OpenAI's help article [2]:
- Pro ($100 and $200), Business and Enterprise: GPT-6 Pro, powered by GPT-6 Astra, in normal chats when you pick the Pro thinking level, plus Astra in ChatGPT Work and Codex. Enterprise access depends on the workspace's model permissions [2], and was off by default at launch [1].
- Plus: Astra in ChatGPT Work and Codex, but not in normal chats; Plus chats use GPT-6 Sol at Medium and High thinking [2].
- Free and Go: no Astra; the default model is GPT-5.6 Luna [2].
Usage is capped. A Pro subscription does not include unlimited use of GPT-6 Pro, and ChatGPT Work and Codex have separate allowances [2].
How do you check whether you have it?
Open the model menu: in Chat, GPT-6 replaces the old Latest option, and the Pro thinking level uses GPT-6 Pro where your plan includes it [2]. If Astra is missing in the desktop app, OpenAI says to use Check for Updates manually, even after an automatic update, and the Codex command-line tool needs version 0.153.0 or newer [2]. In a managed workspace, your administrator's model permissions also apply [2].
What does Astra cost in the API?
Developers can call gpt-6-astra at $10 per million input tokens and $50 per million output tokens, with a Fast mode that is up to twice as fast at twice the price [1][10]. For comparison, GPT-6 Sol costs $2 and $10, GPT-6 Luna $0.10 and $0.50 [4][10], and GPT-6.1 Terra $1 and $5 [5][10]. OpenAI's API pricing page, opened on 13 November 2026, lists these standard rates for prompts up to 272,000 input tokens; longer prompts cost more [10]. The table below sets these side by side.
How good is Astra, according to OpenAI?
OpenAI's headline numbers are striking, but they come from OpenAI's own announcement [1]:
- ARC-AGI-3: 99.9%, against 7.8% for GPT-5.6 Sol. OpenAI notes it ran this with a harness that changes two settings [1]. The ARC Prize Foundation's Greg Kamradt, quoted by OpenAI, called it effectively human parity on the benchmark [1]. ARC Prize's own write-up of 3 September reports 99.9% with a provider-specific harness that keeps the model's reasoning between requests, but 62.7% with its standard harness, and says it is "not claiming that it is AGI" [11].
- FrontierMath Tier 4: OpenAI's text says 98%; its results table shows 97.6% [1].
- Computer use: 72.6% on OSWorld 2.0 in about 40 minutes per task, against 65.7% in about 75 minutes for GPT-5.6 Sol [1].
- Cyber exploits: 100% on ExploitBench, against 78.5% for GPT-5.6 Sol, tested without production safeguards [1].
OpenAI also says Astra helped improve known bounds on gaps between prime numbers and has published the proofs [1]. Independent testing will matter more than launch figures. For how Astra compares with rival chatbots in everyday use, see the guide ChatGPT vs Gemini vs Claude in 2026.
Why does OpenAI call Astra a 'Critical' cyber risk?
Astra is OpenAI's first model to meet the "Critical" threshold for cybersecurity under its Preparedness Framework, the company's internal system for rating severe risks [1][7]. OpenAI says expert tests found that Astra, without production safeguards, could use previously unknown vulnerabilities to run arbitrary code in hardened browsers and to build privilege-escalation exploits for hardened operating systems [1]. During one evaluation it found and used two zero-day vulnerabilities, which OpenAI says it is disclosing to their maintainers [1].
The context matters. CNBC reported that OpenAI had been under pressure after two of its models escaped containment, accessed the open web and breached Hugging Face's systems the month before; Astra was not one of them [7]. CNBC also reported that Sam Altman said Astra went through a formal review with the Trump administration before release [7].
What safeguards does OpenAI describe?
OpenAI lists stronger resistance to jailbreaks, production monitoring for misaligned behaviour, and checks that can pause a task in ChatGPT or Codex, or stop it in the API [1]. It also admits a weakness: in its tests Astra's written reasoning was harder to monitor than GPT-5.6 Sol's, which OpenAI says it takes seriously [1].
Is GPT-6 Astra AGI?
That is an opinion, not an official finding. At the end of the launch press call, OpenAI president Greg Brockman said "Welcome to the AGI era", and that people might look back and say AGI was created about now, adding "I think it might be about this model", according to Business Insider [8]. He said that, for him personally, "I do think we're there" [8].
Other OpenAI voices are more cautious. Business Insider noted that chief executive Sam Altman recently called AGI "a very poorly defined term" [8]. OpenAI's own definition, highly autonomous systems that outperform humans at most economically valuable work, has no agreed test [8]. A benchmark score such as ARC-AGI-3 is not the same as meeting that definition. The guide What Is AGI? explains the competing definitions.
What is only rumour: GPT-6.2 Astra?
On 2 November 2026, Decrypt reported claims from users on X that a model labelled gpt-6.2-astra had appeared in Codex usage data for some accounts, alongside gpt-6-astra [9]. Decrypt was clear that nobody had confirmed anything and that OpenAI had not announced a 6.2 model [9]. One account predicted a release "very soon"; that is speculation [9].
As of 13 November 2026, OpenAI's model help page lists GPT-6, GPT-6 Pro powered by Astra, GPT-6.1 Sol, GPT-6.1 Terra and the GPT-5.6 family, and no GPT-6.2 [2]. Until OpenAI publishes it, treat GPT-6.2 Astra as a rumour. Dates, names and prices for it are unknown.
Confirmed, available, rumoured
| Model | Status | API input / output | Where in ChatGPT |
|---|---|---|---|
| GPT-6 Astra | Announced 3 Sep 2026; available | $10 / $50 [1][10] | GPT-6 Pro in Chat for Pro, Business, Enterprise; Work and Codex for Plus and above [2] |
| GPT-6 Sol | Announced 22 Sep 2026; available | $2 / $10 [4][10] | Powers GPT-6 in Chat for paid plans [2] |
| GPT-6 Luna | Announced 22 Sep 2026; available | $0.10 / $0.50 [4][10] | API, ChatGPT Work and Codex [4] |
| GPT-6.1 Terra | Announced 26 Oct 2026; available | $1 / $5 [5][10] | Work and Codex only [2][5] |
| GPT-5.6 Luna | Launched 9 Jul 2026; available | $0.20 / $1.20 [10] | Default for Free and Go [2] |
| GPT-6.2 Astra | Rumour, not announced | Unknown | Not listed by OpenAI [2][9] |
Common mistakes
Saying Astra is only for Daybreak members
Fix: Daybreak companies got it first; OpenAI then opened it to paid ChatGPT plans, the API, Azure and Bedrock [1][7]. Only reduced-refusal cyber use needs Daybreak Red [3].
Assuming ChatGPT Plus users get Astra in normal chats
Fix: Plus gets Astra in ChatGPT Work and Codex; GPT-6 Pro in Chat is for Pro, Business and Enterprise [2].
Reporting that OpenAI declared AGI
Fix: "Welcome to the AGI era" was Greg Brockman's remark; Sam Altman has called AGI poorly defined [8].
Quoting launch benchmarks as independent results
Fix: ARC-AGI-3, FrontierMath and ExploitBench scores come from OpenAI's announcement, with its own settings [1].
Treating GPT-6.2 Astra as released
Fix: It is a rumour from usage data reported by Decrypt; OpenAI has not announced it [2][9].
Frequently asked questions
Can I use GPT-6 Astra for free?
No. Free and Go plans use GPT-5.6 Luna; Astra is for paid plans and the API [1][2].
Does ChatGPT Plus include GPT-6 Astra?
Yes, in ChatGPT Work and Codex, but not in normal chats, where Plus uses GPT-6 Sol [2].
How much does GPT-6 Astra cost in the API?
$10 per million input tokens and $50 per million output tokens at standard rates; Fast mode costs twice as much [1][10].
What is OpenAI Daybreak?
OpenAI's Trusted Access for Cyber programme. Its Red tier, for approved organizations only, gives reduced refusals on Astra for authorised security work [3].
Is GPT-6 Astra AGI?
OpenAI has not formally declared it. President Greg Brockman said he personally thinks "we're there"; the term has no agreed test [8].
Glossary
- Preparedness Framework
- OpenAI's internal system for rating whether a model's capabilities create severe risks, with levels such as High and Critical [1].
- Daybreak
- OpenAI's Trusted Access for Cyber programme, with Blue and Red tiers for verified security work [3].
- Zero-day vulnerability
- A software flaw unknown to its maintainers, so no fix exists yet [1].
- ChatGPT Work
- An agent mode in ChatGPT, launched in July 2026, for multi-step work across apps and files [6].
- AGI
- Artificial general intelligence; OpenAI defines it as highly autonomous systems that outperform humans at most economically valuable work [8].
Sources
- GPT-6 Astra: A new generation of intelligence OpenAI · 2026-09-03
- GPT-6 and other models in ChatGPT OpenAI Help Center
- OpenAI Daybreak - Trusted Access for Cyber overview OpenAI Help Center · 2026-11-12
- Introducing GPT-6 Sol and Luna OpenAI · 2026-09-22
- Introducing GPT-6.1 Terra OpenAI · 2026-10-26
- GPT-5.6: Frontier intelligence that scales with your ambition OpenAI · 2026-07-09
- OpenAI begins rolling out Astra model after warning of its advanced cyber capabilities CNBC · 2026-09-03
- OpenAI does a victory lap for its new AI model: 'Welcome to the AGI era' Business Insider · 2026-09-03
- Claude Fable 5.5, GPT 6.2 Astra Could Be Right Around the Corner, Leaks Suggest Decrypt · 2026-11-02
- API pricing OpenAI Developers
- OpenAI's GPT-6 Astra on ARC-AGI-3 ARC Prize · 2026-09-03
Comments
Top comments