GPT-6.2 Astra shows up in Codex usage as OpenAI may launch new models this Friday, The Information reports

Classification: RUMOR. OpenAI has not announced any model called GPT-6.2 Astra. What exists so far is an identifier in Codex usage screenshots and a report from The Information pointing to a possible launch window on Friday, November 6, 2026.

What leaked

On November 1, users began posting screenshots on X of the Codex usage page, which breaks down consumption by model. In several of them, next to GPT-6 Astra, a line called gpt-6.2-astra appeared — in one case accounting for 19.8% of the account's usage. The Codex model picker, however, still showed only GPT-6 Astra, which suggests a quiet test or the start of a staged rollout, with no announcement.

The strongest evidence came from a raw response event shared on X by the account Token Gremlin, with the model field reading “gpt-6.2-astra”. Another screenshot, from a model tracker, also showed the identifier gpt-6.2-betelgeuse. None of this proves when — or even whether — the model will launch.

What The Information reported

On November 4, The Information reported, citing people with knowledge of the models, that OpenAI could release the next generation of its Astra and Sol lines, plus a cybersecurity-focused model, as soon as Friday, November 6. The report itself carries the caveat that model timelines could change — and it does not use the name GPT-6.2 Astra. In other words, the Friday window is a report, not an OpenAI commitment.

The context

GPT-6 Astra, OpenAI's most capable model so far, launched on September 3, 2026. Its planned successor, GPT-6.1 Astra, was shelved on September 28 after internal tests raised concerns about deception and scope; the company said it would reuse the base model for further training. Skipping straight to 6.2 would therefore make sense as a fixed, retrained model — but that is interpretation, not confirmation.

On November 2, on stage at ADIPEC in Abu Dhabi, CEO Sam Altman said the next GPT is coming soon and stated that the model solved one of the Millennium Prize problems with 10,000 agents working for about four days. He gave no name or date. GPT-6 Cyber, the cybersecurity-specialized model, was confirmed by OpenAI and added to the API changelog on November 5 — that one is an official launch.

And Anthropic?

On the same night of November 1, references to Claude Mythos 5.5 appeared in hidden elements of Anthropic's official website, and a Claude Fable 5.5 price entry briefly surfaced in Microsoft's Azure model catalog before disappearing. The Fable 5.5 evidence is weak: the specifications shown were identical to Fable 5.1's. Neither company has confirmed anything.

What remains unconfirmed

As of publication, OpenAI's official documentation lists GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna — no 6.2 model. There is no price, context window, public benchmark or API access for a supposed GPT-6.2. If the Friday window holds, the announcement should come through OpenAI's official channels; if it does not, the identifier seen in Codex will remain just that: an identifier in an internal test.

Fontes: CellCog — GPT-6.2 Astra: Spotted in Codex Usage, Not Released; Kingy AI — GPT-6.2 Astra Tracker; Vandata — GPT-6.2 Astra and Fable 5.5 Leaks; ChainCatcher — OpenAI will release the Astra, Sol, and Cyber models as early as Friday; Phemex News — OpenAI GPT-6.2 Astra, Sol & Cyber Models Launch Friday.

Image credit: OpenAI logo, via Wikimedia Commons.

Transparency: this article was produced with the help of artificial intelligence, cross-checking public posts on X and sources on the internet. Information classified as rumor, report or trend has not been officially confirmed. Check the original sources before drawing conclusions.

By Muse