OpenAI's new model exposed: GPT-6.1 Astra video sparks speculation, GPT-6.2…

OpenAI's New Model: 12-Trillion-Parameter Monster Exposed | QbitAI Download the App for Exclusive Early Access | October 31, 2026, 12:08:08 from Beijing | Updated October 31, 2026, 12:52:34

Just today, OpenAI's official account suddenly released a mysterious video, and netizens immediately boiled over, with everyone instantly starting to speculate:

This video, it feels... to a large extent...

was generated by GPT-6.1 Astra, the very model that was urgently halted earlier due to safety issues!!

If that's really the case, the next-generation Astra is probably already on the launch countdown. (OMG)

Seeing the model's results, quite a few netizens have already placed their bets in advance, boldly declaring—

GPT-6.1 Astra surpasses Fable 5.5, with! out! a! doubt!

Some netizens even said outright that if this video was definitely generated by Astra, then...

Please, next week, hurry up, quickly, release it. (Warning.jpg)

But this wave of new-model leaks is "not over yet."

Before GPT-6.1 Astra even got an official announcement, GPT-6.2 has actually been dug up by netizens too?!

Indeed, some netizens, from a third-party model-tracking screenshot, dug out an even bigger new name—

GPT-6.2 Betelgeuse.

Many netizens even speculate that it may correspond to the mysterious model Bel exposed earlier, with a parameter scale very likely exceeding 10 trillion and capabilities possibly above GPT-6 Astra.

GPT-6.1 Astra hasn't even officially returned yet, and GPT-6.2 is already starting to emerge.

Late at night, OpenAI suddenly dropped a mysterious video, and netizens instantly began wondering: is a new model about to arrive?

And among all the speculation, the one repeatedly named and singled out is—GPT-6.1 Astra.

This name... is not unfamiliar to everyone, because this model was actually supposed to debut in October. (doge)

In terms of model positioning, Astra was always the highest-positioned flagship model in the OpenAI series.

On September 3 this year, OpenAI launched GPT-6 Astra, with major upgrades in coding, complex reasoning, computer operation, and long-horizon Agent tasks.

According to the original product cadence, the more capable GPT-6.1 Astra should also have taken over the baton in October~

But right at the last moment, as expected, something went wrong again...

At the end of September this year, The Wall Street Journal reported that GPT-6.1 Astra had encountered safety issues in internal safety and alignment tests:

including problems such as unauthorized execution and unauthorized calls to external tools, and sometimes it was even unable to truthfully report to users which operations it had performed.

Because the model's autonomous behavior exceeded the expected safety boundaries, OpenAI ultimately decided to cancel the originally planned October release and continue improving the safety mechanisms.

(Translation: the model is too strong, too capable, and can't be released for the time being.)

However, although the model hasn't been released yet, quite a few clues about GPT-6.1 Astra's capabilities have already emerged.

According to information previously leaked online, the 6.1 Astra model mainly further upgrades "long-horizon autonomous tasks" and "writing ability."

In particular, it improves the AI "slacking off" problem~ (highlight.jpg)

The new model may be better at figuring things out on its own and pushing forward continuously, able to complete complex tasks end-to-end with less human intervention.

And then, just now, when everyone thought they might not see it in October, OpenAI suddenly made a move again.

Late tonight, the official account suddenly released an animated video of about 76 seconds, titled—

"There are many futures. Which do you want to build?"

What's more interesting is that many netizens have already begun speculating: this animation feels like it was generated by GPT-6.1 Astra!!!

If that's really the case, then Fable 5.5, which has been frequently trending recently with its code animations, is probably about to face a new rival...

Moreover, there really aren't many days left before our Tibo father's 28-day Codex plan wraps up.

First doubling the quota for Pro users, then OpenAI suddenly releasing a mysterious video—it's hard not to overthink.

If on the final day they really did save a flagship model as the closing easter egg, what everyone most looks forward to is probably still GPT-6.1 Astra.

GPT-6.2 "Bel" exposed: the rumored 10-trillion-parameter model?

Just as everyone was discussing GPT-6.1 Astra, another new clue also surfaced.

Around October 29, a netizen shared a screenshot from Discord.

The screenshot shows that a bot named Internal Tracker issued a model update notification.

In it, a string of names prominently appeared: gpt-6.2-betelgeuse - INTERNAL.

Below it was also a line: regions: default.

Let me explain: one is that the version number has directly reached GPT-6.2, and the other is the Betelgeuse after it.

English text is being prepared.

Hey, does this name feel a little familiar to everyone?

That's right, in August this year, OpenAI was reported to be developing a super-large model with the internal codename Bel.

At the time, an anonymous leaker claimed that Bel had completed pre-training and was a follow-up version of the earlier internal project Doug.

The most eye-catching part was its parameter scale, which reached more than "10 trillion" parameters.

If this claim is true, Bel's total parameter count would enter a rather exaggerated level...

However, netizens clearly still felt this wave of leaks wasn't explosive enough.

Bel's true identity hasn't even been confirmed, yet people have already arranged its benchmark scores, pricing, and release date in advance.

The latest reports say that GPT-6.2 Bel may debut this week, and may even make it in time for the finale of Tibo's 28-day Codex plan.

A leaker claiming to have inside information even declared that Bel has already scored 72 points on the Artificial Analysis Intelligence Index!!

If the result is true, that would be 13 points higher than the just-crowned Gemini 4 Carbon.

That's not all—even the parameters and API pricing have been leaked along with it:

According to the leak, GPT-6.2 Bel has a parameter scale of up to 12 trillion, with API pricing at $20 per million tokens for input and $100 for output.

One is GPT-6.1 Astra, originally scheduled for October release but urgently halted due to safety issues.

One is the rumored GPT-6.2 Bel, with 12 trillion parameters and benchmark scores that may refresh the leaderboard.

So here's the question, OpenAI: your next model to be released—is it the B generation or the A generation???

"Special statement: The above content (including videos, pictures, and audio if any) is uploaded and published by users of Dafeng Hao, a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."