GPT-6.2 Astra and Sol missed a reported November 6 launch. What’s known

Where GPT-6.2 stands on November 9

It isn’t out. OpenAI hasn’t announced a GPT-6.2 Astra or Sol, published a price or a system card, or said when either will arrive. What changed this week is how much points toward one, and that a launch date reported for Friday, November 6 came and went with only one of the three models it named.

Here is the week, with whose word each item rests on:

Date	What happened	Whose word	From OpenAI
Nov 1	Codex users post usage views listing gpt-6.2-astra	Screenshots from X users	Nothing
Nov 1	Two banked Codex resets: “Banked 1/2” and “Day 28/ Banked 2/2”	Thibault Sottiaux, who leads Codex, on X	Resets, no model
Nov 2	Altman calls the model behind the Navier–Stokes result “the next GPT,” “going to be released soon”	Sam Altman, on stage at ADIPEC	No name or date
Nov 3	A newer internal model scores 0.875 on BenchCAD 2.0, against 0.594 for GPT-6 Astra	OpenAI’s research post	Unnamed model
Nov 4	New Sol, Astra and cyber models “could be coming as soon as Friday”	The Information	No
Nov 4	Google Cloud Console’s quotas page lists gpt-6.2-astra and gpt-6.2-sol, then doesn’t	Screenshots from X users	No
Nov 5	GPT-6 Cyber released	OpenAI	Yes
Nov 7	The delay was over “serious alignment and safety issues”	One X account, no named source	No

What OpenAI has said

Nothing about GPT-6.2 by name. We checked its pages on November 9:

• The API changelog’s newest entry is gpt-6-cyber, on November 5. The model release notes still end with GPT-6 Rosalind Research on October 30.
• The API models page lists gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna and gpt-6-cyber, but no 6.2. Neither does Codex’s model catalog on GitHub, and Codex’s models page still documents codex -m gpt-6-astra.
• Codex CLI 0.167.0, the newest full release, doesn’t mention it.

OpenAI has, though, said three things that bear on it.

Altman’s “soon.” At the ADIPEC conference in Abu Dhabi on November 2, Altman described the agents behind OpenAI’s Navier–Stokes result: “It was 10,000 agents of the model that is going to be released soon, the next GPT, working for about four days, nonstop, collaborating with each other perfectly,” as @deredleritt3r transcribed the video. The National reported the same remark as “an upcoming GPT model.” Altman gave no name and no date. OpenAI’s own Navier–Stokes post of September 8 said it “used an internal model that is significantly more capable than GPT‑6 Astra,” trained “since August 28.”

The engineering score. On November 3, an OpenAI research post said: “A newer internal model achieved an average score of 0.875, compared with 0.594 for GPT‑6 Astra at its highest reasoning setting,” on BenchCAD 2.0, a test of rebuilding 3D parts and reading circuit boards. It doesn’t say whether that’s the Navier–Stokes model, or whether either will be called GPT-6.2.

The Astra it scrapped. On September 28, OpenAI said it wouldn’t release GPT-6.1 Astra, which, its head of safety systems told CNN, “didn’t quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it’s done.” Quartz, citing The Wall Street Journal, says OpenAI intends to take its underlying model through more reinforcement learning “to build out subsequent entries in the GPT-6 family.”

The launch reported for November 6

On November 4, Stephanie Palazzolo of The Information wrote on X: “a scooplet: new Sol, Astra and cyber models from OpenAI could be coming as soon as Friday,” linking to The Information’s paywalled newsletter. Coverage relaying it, such as ChainCatcher’s, says the report gave no model numbers, prices or context windows.

The cyber model came a day early. OpenAI released GPT-6 Cyber on November 5, for authorized security research through its Daybreak program, and rates it Critical in cybersecurity under its Preparedness Framework (our memo covers who can use it). OpenAI didn’t say whether it was one of the models The Information meant.

The Astra and the Sol didn’t come. OpenAI has said nothing about them, so nothing official was delayed. On November 7, an X account called Token Gremlin wrote that “some serious alignment and safety issues were discovered at the last minute and needed to be fixed,” adding: “A few more days of polishing and fixing those issues, and we should finally see the models launch.” It names no source, and we can’t check it. The post had more than 250,000 views by November 9, and no one has confirmed it.

What people have seen

In Codex, since November 1. Users posted usage views naming a model OpenAI hasn’t announced. @leo114119 wrote, in our translation, that although Codex shows GPT-6 Astra, the usage view shows GPT-6.2 Astra; @Mashiro_0619 posted views “collected from three users,” by our translation; and @Mr_Salio wrote: “The UI still GPT-6 Astra but the API response breakdown now shows ‘gpt-6.2-astra’.” Screenshots can be edited, and none of this has been checked against an account.

On Google Cloud, November 4. OpenAI’s models are on Google Cloud in private preview: Google’s announcement of October 21 names GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna “in private preview to select enterprise customers.” On the afternoon of November 4, Pacific time, @lyraxana and @AiBattle_ posted the Google Cloud Console’s quotas page, filtered for “gpt-6.2,” offering two values: base_model:gpt-6.2-astra and base_model:gpt-6.2-sol. AiBattle added, “Release likely tomorrow.” About three hours later, @Mr_Salio posted the same filter showing nothing: “OpenAI has removed GPT-6.2 Astra and GPT-6.2 Sol from the Google Cloud.” A quota entry is a line in a console, not a model anyone could call.

In a terminal, November 6. @Layton_Gott, whose profile describes open-sourcing AI coding tools, posted a cropped Codex screen reading gpt-6.2-astra xhigh, captioned “went down to visit Tibo in the basement and found this lol,” a joke on Sottiaux’s post “I live in the basement now.” Sottiaux replied: “Basement is full of wonders.” A line of terminal text is easy to make, and the reply confirms nothing.

Five names, and what backs each

Name	Where it comes from	What backs it	From OpenAI
“The next GPT”	Altman at ADIPEC, November 2	OpenAI’s CEO, on stage	No name, price or date
gpt-6.2-astra	Codex usage views, November 1; Google Cloud’s quotas page, November 4	Screenshots from several X accounts	Nothing
gpt-6.2-sol	Google Cloud’s quotas page, November 4	Two screenshots of one filter, gone hours later	Nothing
gpt-6.2-betelgeuse	A Discord bot called “Internal Tracker,” posted by @taker_of_whizz, October 29	One screenshot	Nothing
Bel	A post by @synthwavedd, August 25, calling it OpenAI’s next pretraining run, with more than 10 trillion parameters	No named source	Nothing

@ChrisGPT, an account that says it reported on Bel earlier, is careful about tying Bel to Betelgeuse: “Could this be the name change I’d heard about? Possibly. But I have not confirmed yet.”

When it might arrive

OpenAI hasn’t said. Altman’s “soon” is the only timing from the company, and the November 6 date came from The Information’s sources, not from OpenAI.

Polymarket asks when the next Astra numbered 6.1 or higher reaches the general public; its rules count only models with Astra in the name and leave out cyber and Daybreak models, so GPT-6 Cyber didn’t settle it. On the morning of November 9, with about $128,000 traded, it had 61% on November 9 to 15, 15% on the week after, 12% on the week after that, and under 1% on no release by January 31. A separate market on a Sol numbered 6.2 or higher had 66% by November 15 and 82% by November 30. That’s what bettors think, not information.

What to do

1. Don’t buy or change a plan over it. There’s no price and no list of plans. What is confirmed about your limits is elsewhere, such as ChatGPT and Codex sharing one allowance from December.
2. Hold your banked resets, if you can afford to. A banked reset is “saved to your account for later use,” and “you choose when to use it before it expires,” says OpenAI’s help article. In the desktop app, open Settings > Usage & billing to see Available resets and their expirations; in Codex CLI, enter /usage and select Redeem reset. Sottiaux’s two resets on November 1 followed the 28-day pledge and a hint that the next two days would be “resets… banked…,” not a model. If you’re blocked, use one.
3. When it lands, check what it costs before you make it your default. In Codex, GPT-6 Astra costs 250 credits per million input tokens and 1,250 per million output, says Codex pricing. A new Astra may not cost the same.
4. Keep your approvals and sandbox as they are. The Astra OpenAI scrapped in September went out of scope in testing, and the one claim about this delay is about alignment. That’s no finding about GPT-6.2, but a model’s first week is no time to loosen anything.

If GPT-6.2 reaches Codex

Enjoy runs Codex on your ChatGPT plan and asks Codex for its current models each time you open the settings, so a new model appears there without an Enjoy update, once your Codex offers it. If it’s missing, update Codex. If a new model burns through your allowance faster, you can switch the conversation to another agent, such as Claude Code, and it catches up on what happened so far. Enjoy doesn’t add usage to a ChatGPT plan, and it can’t give you a model your account doesn’t have.

What OpenAI hasn’t said

• Whether GPT-6.2 exists, under that name or another.
• Whether “the next GPT” is the Navier–Stokes model, the BenchCAD model, or both, and whether it’s the reworked GPT-6.1 Astra.
• Why the Astra and the Sol didn’t come with GPT-6 Cyber, or whether they were ever meant to.
• What gpt-6.2-astra in Codex usage is, and whether users are told when a request runs on something other than the model the picker names.
• Any price, plan, date or system card for a GPT-6.2.

We’ll update this memo when OpenAI announces it, or says why it hasn’t.