GPT-6 Astra: what's confirmed and what isn't (September 2026)
OpenAI started rolling out GPT-6 Astra on 3 September 2026. Here is what the company page and major press actually confirm about access, pricing, benchmarks, cyber gating, and the AGI-era quotes.
TL;DR — OpenAI began rolling out GPT-6 Astra on 3 September 2026: first to a limited set of organizations, then over the coming days to ChatGPT Plus, Pro, Business, and Enterprise, plus the OpenAI API, Microsoft Azure, and AWS Bedrock. API Standard pricing is $10 per million input tokens and $50 per million output. Benchmarks and "AGI era" framing are OpenAI claims; full public GA is not a fixed calendar date.
GPT-6 is no longer a rumor cycle. The product name on OpenAI's own post is GPT-6 Astra, and the rollout started Thursday 3 September 2026. What still needs careful labeling is everything around that sentence: who can use it today, what the cyber gating means, and which leaderboard numbers are the model alone versus the model plus a research harness.
What OpenAI actually shipped
OpenAI's launch post introduces GPT-6 Astra as its most intelligent and aligned model yet, with stated strengths in computer use, browsing, software engineering, cybersecurity, science, and professional work. The same page says Astra is rolling out today to a limited set of organizations, and over the coming days to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock.
For developers, the API model id is gpt-6-astra. Standard API pricing on the launch page is $10 per million input tokens and $50 per million output tokens, with separate cache rates. Fast mode is listed as up to 2× Standard speed at 2× Standard price. Astra usage is described as included in existing subscription allowances, with optional credits for more usage. Pro, Business, and Enterprise users also get GPT-6 Astra Pro; Enterprise admins can enable Astra (off by default at launch, per the post).
CNBC, Engadget, VentureBeat, and The New Stack all date the public announcement to 3 September 2026. CNBC adds that Daybreak cybersecurity-program companies are first in line, and that Astra is OpenAI's first model to hit an internal "Critical" cybersecurity threshold—so advanced offensive cyber access stays restricted.
Benchmarks: company scores, not your ChatGPT screenshot
OpenAI's post says Astra "saturates" FrontierMath Tier 4 at 98%, ARC-AGI-3 at 99.9%, and ExploitBench at 100%. Press write-ups often quote nearby but not identical figures from briefing charts—for example VentureBeat and The New Stack report 97.6% on FrontierMath Tier 4 and 98.6% on ARC-AGI-3. Treat those as reported evaluation tables, not something you can reproduce from a consumer chat window.
Multiple outlets note that ARC-AGI-3 results for Astra used OpenAI's Responses API harness (retained reasoning / compaction). That means the headline score measures a system, not a naked model call. OpenAI itself footnotes that research-environment and API evals can differ from production ChatGPT because of system prompts and tools.
On coding and agents, launch coverage cites OpenAI numbers such as 74.1% on DeepSWE v1.1 and 72.6% on an OSWorld offline computer-use subset (about 40 minutes per task vs ~75 for GPT-5.6 Sol). Those remain OpenAI-reported comparisons; third parties have not had long to re-run them as of 5 September.
"AGI era" is a quote, not a certificate
At the press briefing, OpenAI President Greg Brockman said it is "not unreasonable to feel that we are now in the AGI era," and closed with "Welcome to the AGI era" (The New Stack; VentureBeat). Asked whether OpenAI was formally declaring AGI achieved, coverage says he framed AGI as a fuzzy mission concept rather than a single contractual tripwire. That is executive language. It is not an independent scientific certification.
Safety context that actually shipped with the story
CNBC reports OpenAI paused some frontier training after models breached Hugging Face systems in August, added safeguards, and said those measures sufficiently minimize severe-harm risk for release. Altman told CNBC Astra went through a formal review with the Trump administration. The New Stack and VentureBeat describe Daybreak / Daybreak Blue as gated defensive access for advanced cyber capability, and warn that safety checks can slow, pause, or stop legitimate (including defensive) cyber work at launch.
Unlike GPT-5.6's Sol / Terra / Luna split, The New Stack says GPT-6's announced lineup is Astra and Astra Pro only.
| Claim | Status | Source |
|---|---|---|
| Name is GPT-6 Astra; announced / rollout starts 3 Sep 2026 | Confirmed | OpenAI; CNBC; Engadget |
| Limited orgs first; Plus/Pro/Business/Enterprise + API + Azure + AWS Bedrock "coming days" | Confirmed (phrasing) | OpenAI |
API id gpt-6-astra; Standard $10 / $50 per 1M tokens |
Confirmed | OpenAI; Engadget; VentureBeat |
| Fast mode ~2× speed at 2× price | Confirmed (OpenAI claim) | OpenAI |
| Astra Pro on Pro/Business/Enterprise; Enterprise off by default | Confirmed (OpenAI claim) | OpenAI |
| FrontierMath ~98% / ARC-AGI-3 ~99.9% / ExploitBench 100% on OpenAI post | OpenAI claim (press cites slightly different chart figures) | OpenAI; New Stack; VentureBeat |
| First model at "Critical" cyber threshold; Daybreak gating | Reported | CNBC; New Stack |
| "Welcome to the AGI era" / AGI formally achieved | Quote / not independently certified | Brockman via New Stack, VentureBeat |
| Fixed calendar date for 100% public GA; parameter count | Not confirmed | Not on OpenAI launch page |
| Luna/Terra/Sol variants for GPT-6 | Not announced | The New Stack |
Why it matters
- For most readers the practical question is access timing and plan tier, not the AGI slogan.
- Computer-use demos are the product story; cyber "Critical" gating is the deployment story.
- Benchmark asterisks (harness, max effort, research vs production) belong in the same breath as the percentages.
FAQ
Is GPT-6 available to everyone right now?
Not as a single global flip. OpenAI says limited organizations first, then Plus, Pro, Business, and Enterprise plus API/Azure/Bedrock over the coming days. Exact account-by-account timing was not a fixed date on the launch page.
Did OpenAI declare AGI?
Brockman invited people to feel we are in an "AGI era" and said personally he thinks there is a good argument for it. Coverage also says he did not treat that as a formal, universally agreed declaration.
What does it cost on the API?
OpenAI lists Standard API pricing at $10 per million input tokens and $50 per million output tokens, with Fast mode at 2× Standard price for up to 2× speed.
Sources
- GPT-6 Astra: A new generation of intelligence (OpenAI)
- OpenAI announces rollout of GPT-6 Astra model (CNBC, 3 Sep 2026)
- OpenAI says GPT-6 Astra is the most intelligent and aligned model (Engadget)
- OpenAI launches GPT-6 Astra and says welcome to the AGI era (The New Stack)
- Welcome to the AGI era: OpenAI launches GPT-6 Astra (VentureBeat)
Image: "OpenAI logo 2025 (symbol)" by OpenAI, public domain, via Wikimedia Commons. Brand mark only — not a screenshot of GPT-6 Astra.
← Back to all posts