Independent newsroom

Thursday, September 10, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

OpenAI · Thursday, September 10, 2026

OpenAI starts rolling out GPT-6 Astra

The File desk · Sep. 10, 2026, 10:18 a.m. ET

Confirmed4 primary sources

Status

Confirmed from OpenAI’s GPT-6 Astra launch post (live openai.com returned 403 this pass; Internet Archive capture retrieved) and from the Deployment Safety Hub system card, whose changelog was updated Sept. 9, 2026. Company scores are OpenAI’s. Outside labs have not been independently verified here as accepting every claimed score.
  • Confirmed
    OpenAI’s launch post says GPT-6 Astra is rolling out first to a limited set of organizations, then over coming days to ChatGPT Plus, Pro, Business, and Enterprise users, plus the OpenAI API, Microsoft Azure, and Amazon Bedrock. The API model id is gpt-6-astra. Standard API list price is $10 per million input tokens and $50 per million output tokens. Pro, Business, and Enterprise plans also get GPT-6 Astra Pro. Enterprise admins must turn Astra on; it starts off. The company says Astra meets the Critical cybersecurity threshold under its . The system card changelog was updated Sept. 9, 2026.
  • Unverified
    OpenAI’s published scores — including FrontierMath Tier 4 at 98 percent, ARC-AGI-3 at 99.9 percent, ExploitBench at 100 percent, and OSWorld 2.0 latency figures — are the company’s. Greg Kamradt’s ARC Prize Foundation comment is quoted on OpenAI’s page. Outside labs have not been independently verified here as accepting every claimed score. Live openai.com pages returned 403 to this desk.
OpenAI is putting GPT-6 Astra in more people’s hands. The company’s launch post says the model is rolling out first to a limited set of organizations, then over coming days to ChatGPT Plus, Pro, Business, and Enterprise users, plus the OpenAI API, Microsoft Azure, and Amazon Bedrock. The API model id is gpt-6-astra. List price on the standard API tier is $10 per million input tokens and $50 per million output tokens. Pro, Business, and Enterprise plans also get GPT-6 Astra Pro. Enterprise admins must turn Astra on; it starts off. OpenAI calls Astra its most capable model for broad deployment and says it is state of the art on computer use, browsing, software engineering, cybersecurity, science, and professional work. Company scores include FrontierMath Tier 4 at 98 percent, ARC-AGI-3 at 99.9 percent, and ExploitBench at 100 percent. On OSWorld 2.0 latency tests, Astra scored 72.6 percent in about 40 minutes per task, versus 65.7 percent in about 75 minutes for GPT-5.6 Sol. Greg Kamradt of the ARC Prize Foundation said Astra reached human parity on ARC-AGI-3 action efficiency on 96 percent of levels. The company says Astra meets the Critical cybersecurity threshold under its . A system card at deploymentsafety.openai.com/gpt-6-astra — changelog updated Sept. 9 — and a safety overview accompany the launch. Live openai.com pages returned 403 this pass. Outside labs have not been independently verified here as accepting every claimed score.
OpenAI is putting GPT-6 Astra in more people’s hands.
The company’s launch post says the model rolls out first to a limited set of organizations. Over coming days it is to reach ChatGPT Plus, Pro, Business, and Enterprise users, and the OpenAI API, Microsoft Azure, and Amazon Bedrock.
The API model id is gpt-6-astra. On the standard API tier the list price is $10 per million input tokens and $50 per million output tokens. Pro, Business, and Enterprise plans also get GPT-6 Astra Pro. Enterprise administrators can turn Astra on; access starts off.
OpenAI calls Astra its most capable model for broad deployment. It says Astra is state of the art on computer use, browsing, software engineering, cybersecurity, science, and professional work.
Company scores on that page include FrontierMath Tier 4 at 98 percent, ARC-AGI-3 at 99.9 percent, and ExploitBench at 100 percent. On OSWorld 2.0 latency tests, OpenAI says Astra scored 72.6 percent in about 40 minutes per task, against 65.7 percent in about 75 minutes for GPT-5.6 Sol.
Greg Kamradt of the ARC Prize Foundation, quoted on OpenAI’s page, said Astra “surpassed our human action-efficiency baseline on 96% of levels, effectively reaching human parity on the benchmark.”
OpenAI says Astra meets the Critical cybersecurity threshold under its . The system card on the Deployment Safety Hub has a changelog dated Sept. 9, 2026. A safety overview sits at openai.com/index/safety-overview-gpt-6-astra/.
Live openai.com pages returned 403 to this desk. The launch-post text is from an Internet Archive capture of the same URL. Outside labs have not been independently verified here as accepting every claimed score. Anthropic’s newsroom still leads with Claude Fable 5.1 and Claude Mythos 5.1 from Sept. 1.
How we know

What is still unknown or disputed

  • Company scores are OpenAI’s; outside labs have not been independently verified here as accepting every claimed score.
  • Live openai.com/index/gpt-6-astra/ and openai.com/index/safety-overview-gpt-6-astra/ returned 403 this pass.

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1 · fetch incomplete

    GPT-6 Astra: A new generation of intelligence

    OpenAI · September 3, 2026

    Live page returned HTTP 403. See Internet Archive capture of the same URL.

    https://openai.com/index/gpt-6-astra/

  2. Source 2

    GPT-6 Astra: A new generation of intelligence (Internet Archive capture)

    Internet Archive / OpenAI · September 3, 2026

    Rolling out first to a limited set of organizations, then Plus, Pro, Business, Enterprise, API, Azure, and Bedrock. Model id gpt-6-astra. Standard API $10 / $50 per million tokens. FrontierMath Tier 4 98%; ARC-AGI-3 99.9%; ExploitBench 100%. OSWorld 2.0: 72.6% in about 40 minutes vs 65.7% in about 75 minutes for GPT-5.6 Sol. Enterprise access off by default.

    https://web.archive.org/web/2026/https://openai.com/index/gpt-6-astra/

  3. Source 3

    GPT-6 Astra System Card

    OpenAI Deployment Safety Hub · September 9, 2026

    Changelog September 9, 2026: updates to the Alignment section. Card states Astra is the most capable model OpenAI has broadly deployed and meets the Critical cybersecurity level under the .

    https://deploymentsafety.openai.com/gpt-6-astra

  4. Source 4 · fetch incomplete

    Safety overview: GPT-6 Astra

    OpenAI · checked 2026-09-10

    Live page returned HTTP 403 this pass.

    https://openai.com/index/safety-overview-gpt-6-astra/