Independent newsroom

Monday, September 14, 2026

The File

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

OpenAI · Monday, September 14, 2026

OpenAI says Perplexity is using GPT-6 Astra on live systems

The File desk · Sep 14, 2026, 10:22 AM UTC

Status

Confirmed from OpenAI’s customer story dated Sept. 14, 2026. Live openai.com returned 403 this pass. This is a customer/engineering write-up, not a new frontier model launch. Anthropic’s newsroom, fetched live, still led with its Sept. 10 threat report and the Sept. 1 Claude Fable 5.1 / Mythos 5.1 announcement. Live x.ai/news returned 403; a Sept. 9 archive listed nothing newer than Sept. 4. Live Meta AI blog returned 400. Google DeepMind’s public index still listed September WeatherNext and Gemini 3.8 Flash items.

  • Confirmed

    OpenAI published a customer story dated Sept. 14. The post says Perplexity uses GPT-6 Astra to write communications, change software, and monitor production systems, and that staff check in less often than with earlier models. Johnny Ho, Perplexity’s cofounder and chief strategy officer, is quoted saying the model can craft communications, edit real-world systems, and monitor production software in ways earlier generations could not. The piece describes asking Astra to build small testing programs that stand in for other services so workflows can be checked end to end.

  • Unverified

    Independent measurement of how often Perplexity staff intervene, and of production reliability under Astra, is not available beyond the OpenAI post and quoted remarks. Live openai.com returned 403.

OpenAI published a customer story dated Sept. 14. The post says Perplexity uses GPT-6 Astra to write communications, change software, and monitor production systems, and that staff check in less often than with earlier models. Johnny Ho, Perplexity’s cofounder and chief strategy officer, is quoted saying the model can craft communications, edit real-world systems, and monitor production software in ways earlier generations could not. The piece describes asking Astra to build small testing programs that stand in for other services so workflows can be checked end to end. This is a customer/engineering write-up, not a new frontier model launch. Anthropic’s newsroom still led with its Sept. 10 threat report and the Sept. 1 Claude Fable 5.1 / Mythos 5.1 announcement. xAI’s news page showed nothing newer than early September in this window. Live Meta’s AI blog returned 400 this pass. Google DeepMind’s public index listed September WeatherNext and Gemini 3.8 Flash items from early September, not overnight.

OpenAI put up a customer story dated Sept. 14.

The post says Perplexity uses GPT-6 Astra to write communications, change software, and monitor production systems. It says staff check in less often than they did with earlier models.

Johnny Ho, Perplexity’s cofounder and chief strategy officer, is quoted: “We can have the model craft communications, edit real-world systems, and monitor our production software in a way that previous generations were not able to.”

The piece describes asking Astra to build small testing programs that stand in for other services — for example a language-model interface or a connector — so a workflow can be checked from start to finish.

Ho is also quoted: “We’re actually able to trust it with full end-to-end systems and check in on it much less frequently than previous generations of models.”

This is a customer write-up. It is not a new frontier model launch.

Live openai.com returned 403 to this environment. Wayback holds Sept. 12 captures of the same URL that did not render the article body here.

Anthropic’s newsroom, fetched Monday, still listed the Sept. 10 threat report and the Sept. 1 Claude Fable 5.1 and Claude Mythos 5.1 announcement. There was no newer frontier-model launch on that page.

Live x.ai/news returned 403. An Internet Archive capture from Sept. 9 still showed dated posts through Sept. 4 (Grok Bot items). This desk could not read a later live xAI listing.

Live ai.meta.com/blog returned 400. Google DeepMind’s public index still listed Gemini 3.8 Flash and WeatherNext under September 2026. Nothing newer overnight was on that page.

What is still unknown or disputed

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

    Perplexity trusts GPT-6 Astra with end-to-end systems

    OpenAI · September 14, 2026

    Sept. 14, 2026. Perplexity uses Astra to write communications, change software, and monitor production; Johnny Ho quotes on end-to-end systems and less frequent check-ins. Live page returned HTTP 403 this environment.

    https://openai.com/index/perplexity-improving-accuracy-with-astra/

  2. Source 2

    Anthropic Newsroom (checked)

    Anthropic · checked 2026-09-14

    Newest listed remain Sept. 10 threat report (Detecting and countering misuse of AI: September 2026) and Sept. 1 Fable/Mythos 5.1; no newer frontier launch overnight.

    https://www.anthropic.com/news

  3. Source 3

    xAI News (live page blocked; Sept. 9 archive checked)

    Internet Archive / xAI · checked 2026-09-14

    Live https://x.ai/news returned HTTP 403. Archive capture 20260909130021 lists dated posts through Sept. 4, 2026 (Grok Bot items Sept. 3–4). No Sept. 13–14 items on that capture.

    https://web.archive.org/web/20260909130021/https://x.ai/news

  4. Source 4

    Google DeepMind News (checked)

    Google DeepMind · checked 2026-09-14

    Public index listed Gemini 3.8 Flash and 3.8 Flash Cyber under September 2026 Models, plus WeatherNext among science items. No newer overnight frontier launch on the fetched page.

    https://deepmind.google/blog/

  5. Source 5 · fetch incomplete

    Meta AI blog (checked)

    Meta · checked 2026-09-14

    Live page returned HTTP 400 this pass. No Meta launch is invented here.

    https://ai.meta.com/blog/