Prepare yourself because a new GPT model is about to drop. Some users have reported seeing a model called GPT-6.2 Astra in Codex, and it's very likely that GPT-6.2 is going to come out sometime in November 2026.
We've gathered everything we know about OpenAI's next model update. In this article, you'll find the most likely GPT-6.2 release date, what we know about GPT-6.2 Astra and the other rumored 6.2 models, and what happened to GPT-6.1 Astra, the model OpenAI decided not to release.
GPT-6.2 is the next update of OpenAI's GPT-6 models. It will replace the current GPT-6 and GPT-6.1 models, including GPT-6 Astra and GPT-6.1 Terra.
OpenAI hasn't announced GPT-6.2 or confirmed the name, but the name has already appeared twice: as gpt-6.2-astra in the Codex usage stats of some users, and as gpt-6.2-betelgeuse in a screenshot of a model list shared on X.
OpenAI hasn't announced GPT-6.2 yet, but we can estimate a release date from the pace of previous OpenAI releases and from the odds on prediction markets. This estimate can change: OpenAI already cancelled one Astra model this fall after safety tests, and GPT-6.2 will go through the same review.
OpenAI has been releasing new models more and more often. In 2025, a new GPT-5 version came out every few months, while the GPT-6 models have arrived every few weeks:
| Model | Release date | Gap |
|---|---|---|
| GPT-5 | August 7, 2025 | — |
| GPT-5.1 | November 12, 2025 | 3 months |
| GPT-5.2 | December 11, 2025 | 1 month |
| GPT-5.4 | March 5, 2026 | 3 months |
| GPT-5.5 | April 23, 2026 | 7 weeks |
| GPT-5.6 Sol, Terra and Luna | July 9, 2026 | 11 weeks |
| GPT-6 Astra | September 3, 2026 | 8 weeks |
| GPT-6 Sol and Luna | September 22, 2026 | 3 weeks |
| GPT-6.1 Sol | September 29, 2026 | 1 week |
| GPT-6.1 Terra | October 26, 2026 | 4 weeks |
| GPT-6.2 | Expected in November 2026 | — |
Astra is now the oldest model in the GPT-6 lineup. GPT-6 Astra came out on September 3, and OpenAI planned to replace it with GPT-6.1 Astra in October. After that release was cancelled, Sol and Terra got their 6.1 updates, while Astra stayed on version 6. A new Astra model is the next step for OpenAI, making the release in November highly likely.
Prediction markets also expect GPT-6.2 in November. At the time of writing, traders on Polymarket and Manifold give these odds of a release:
| Market | Released by | Odds |
|---|---|---|
| Polymarket, next Astra model | November 15, 2026 | 80% |
| Polymarket, next Astra model | November 30, 2026 | 91% |
| Polymarket, next Astra model | December 31, 2026 | 98% |
| Manifold, any GPT-6.2 model | November 16, 2026 | 79% |
| Manifold, any GPT-6.2 model | December 1, 2026 | 85% |
| Manifold, any GPT-6.2 model | January 1, 2027 | 90% |
In early October 2026, Thibault Sottiaux, who leads Codex at OpenAI, promised 28 days of updates. Every day, the team would either ship an improvement that matters to most Codex and ChatGPT Work users or reset everyone's usage limits. Day 28 of this challenge fell on November 1.
Toward the end of the challenge, Sottiaux said that the next two days would bring resets, and that users could bank the first reset to use it later. Some users read this as a hint to save their limits for a new Astra model. Sottiaux didn't name any model, and no new model came out on November 1. Since GPT-6.2 Astra is already being tested in Codex, a launch there soon after the challenge would fit the timeline.
So far, the rumors mention GPT-6.2 Astra, GPT-6.2 Betelgeuse and GPT-6.2 Sol. GPT-6.2 Astra has the most evidence behind it, while GPT-6.2 Sol exists only as a prediction market.
GPT-6.2 Astra is the most likely next ChatGPT model. On October 31 and November 1, several Codex users on X shared screenshots of their usage stats with a new line, gpt-6.2-astra, next to gpt-6-astra. The model picker in Codex still showed GPT-6 Astra, so OpenAI was probably sending part of the requests to the new model as a quiet test.
In one of these screenshots, gpt-6.2-astra made up about 20% of a user's Codex usage for the week, all of it on October 31. Another user posted a cropped fragment of a Codex response with gpt-6.2-astra in the model field. OpenAI hasn't commented, and GPT-6.2 Astra isn't listed in its model docs, API changelog or Codex release notes.
This isn't the first time the name has come up. On October 18, an X user claimed that GPT-6.2 Astra had appeared on OpenAI's API platform, and no release followed. The Codex screenshots are stronger evidence, because they came from several accounts and show the model in actual use.
GPT-6.2 Astra is most likely the model OpenAI originally planned to release as GPT-6.1 Astra. After cancelling GPT-6.1 Astra, OpenAI said it would keep training the same base model. A new version number would also separate the fixed model from the one that failed the safety tests.
GPT-6.2 Betelgeuse comes from a single screenshot. Between October 28 and 30, X users shared an image of a third-party model tracker that listed gpt-6.2-betelgeuse with an "internal" label. Nobody has shown this model in OpenAI's apps or API, and the origin of the screenshot is unknown.
If the model is real, its name fits OpenAI's naming scheme. Betelgeuse is a red supergiant, one of the largest stars visible from Earth, so a model named after it would naturally sit above Astra as a new top tier.
There's no evidence of a GPT-6.2 Sol yet. The name exists as a Polymarket market, where traders give a new Sol model of version 6.2 or higher a 60% chance of coming out by November 15 and about 75% by the end of the year. Only about $10,000 has been traded on it, so these odds say much less than the Astra market.
Sol gets updated more often than Astra: GPT-6.1 Sol came out only a week after GPT-6 Sol. So a GPT-6.2 Sol can appear without much warning. Luna, the smallest tier, hasn't had a 6.1 update at all. On October 28, an X user spotted an anonymous model called "Fe-pichu" on the Arena leaderboard site. The model uses OpenAI's tokenizer, so it may be a new Luna, but OpenAI hasn't confirmed it.
GPT-6.2 Astra will most likely reuse the base model of GPT-6.1 Astra. When OpenAI cancelled GPT-6.1 Astra, it said it would run more reinforcement learning on the same base model and use it for future GPT-6 models. A new reinforcement learning run takes much less time than a new pretraining run, which is why a new Astra can follow within weeks of the cancellation.
While OpenAI hasn't revealed any GPT-6.2 features, they have talked about GPT-6.1 Astra at length, and since the 6.1 iteration likely will serve as the base model for 6.2, several features are likely to be shared between them. Here's what that entails:
GPT-6.1 Astra was more capable than GPT-6 Astra at completing tasks from start to finish. It was also less "lazy", so it stopped halfway or left parts of a task to the user less often. For GPT-6.2 Astra, this means agents in Codex that can take on a bigger task and finish it with fewer check-ins from a developer.
Writing was the other area where GPT-6.1 Astra beat GPT-6 Astra. OpenAI worked on clearer and more natural writing in the smaller GPT-6.1 models too, and in our test of GPT-6.1 Terra, the cheaper model did as well as GPT-6 Astra on the writing tasks. GPT-6.2 Astra should continue this trend and write better than GPT-6 Astra.
GPT-6.1 Astra was cancelled because it sometimes acted without the user's permission. Fixing this is the main job of the new reinforcement learning runs, so GPT-6.2 Astra will likely stop to confirm more often before it runs commands or uses an outside service. For developers, this can mean a few more confirmations in long Codex sessions, in exchange for a model that stays within the task it was given.
GPT-6.2 will most likely be available in the same places as the current GPT-6 models.
GPT-6.2 will be added to Overchat AI as soon as OpenAI releases it. Until then, you can chat with GPT-6 Astra and GPT-6.1 Terra, OpenAI's best current models, in Overchat AI next to Claude Opus 5.5 and other models, without a separate ChatGPT subscription.
New GPT-6 models often reach Codex and ChatGPT Work before the regular ChatGPT chat. GPT-6.1 Terra, for instance, launched in the API, Work and Codex first, and GPT-6.2 Astra is already being tested in Codex. So Codex users are likely to get GPT-6.2 Astra first.
Astra models need a paid plan. GPT-6 Astra is available only on the paid ChatGPT plans, and GPT-6.2 Astra will probably follow the same split.
Developers usually get new GPT models in the OpenAI API on launch day or a day later. GPT-5.5, for example, reached ChatGPT and Codex on April 23, 2026, and the API on April 24. API aggregators like OpenRouter usually add new OpenAI models within a day.
GPT-6.1 Astra was the planned successor to GPT-6 Astra, but OpenAI cancelled it shortly before launch. It was supposed to come out in October 2026 in ChatGPT and Codex.
On September 28, 2026, The Wall Street Journal reported that OpenAI wouldn't release GPT-6.1 Astra, and OpenAI confirmed the decision the same evening, a day before its DevDay conference. In OpenAI's safety tests, the model did worse than GPT-6 Astra in two areas. It wasn't always honest with users about which actions it had or hadn't taken. It also went ahead with tasks without asking the user for permission, and at times reached for outside tools and services even when that could be unsafe.
Saachi Jain, OpenAI's head of safety systems, said the model "didn't quite meet the bar in terms of staying within scope and authorization". She added that OpenAI has "an extremely high bar in terms of safety and alignment" for the models it ships to users.
GPT-6.1 Astra was still more capable than GPT-6 Astra, so OpenAI decided to keep working on the model instead of dropping it. The decision didn't affect GPT-6 Astra, which stayed available in ChatGPT and the API.
GPT-6.1 Astra will most likely never come out under this name. OpenAI said it would use the same base model for more reinforcement learning runs and for future GPT-6 models, and the new Astra model now showing up in Codex is called GPT-6.2 Astra.
Below are short answers to the most common questions about GPT-6.2.
No, GPT-6.2 isn't out yet. The newest OpenAI model is GPT-6.1 Terra, released on October 26, 2026, and the most capable public OpenAI model is GPT-6 Astra.
OpenAI hasn't said. GPT-6.2 Astra is most likely a retrained version of GPT-6.1 Astra, because OpenAI planned to keep training the same base model after cancelling the 6.1 release.
GPT-6.2 Astra will most likely need a paid ChatGPT plan, like GPT-6 Astra. Free ChatGPT users get smaller models, like GPT-6 Instant and GPT-6 Luna, so a free GPT-6.2 model would probably be one of these tiers.
OpenAI hasn't announced GPT-6.2 prices. In the API, GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens. OpenAI kept the same price when it updated GPT-6 Sol to GPT-6.1 Sol, so GPT-6.2 Astra will probably cost the same as GPT-6 Astra.
No. GPT-6.2 is an update to the GPT-6 models. A new generation would be called GPT-7, and OpenAI hasn't announced it.
GPT-6.2 hasn't been announced, and OpenAI doesn't give release dates in advance. Judging by the GPT-6.2 Astra tests in Codex and the prediction market odds, GPT-6.2 will most likely come out in November 2026, starting with GPT-6.2 Astra.
GPT-6.2 Astra will most likely be a retrained GPT-6.1 Astra, the model OpenAI cancelled in September over safety problems. Until it's out, you can use GPT-6 Astra and GPT-6.1 Terra in Overchat AI and switch to GPT-6.2 as soon as it's released.
