Short answer
GPT-6.2 Astra has not been released. As of November 3, 2026, OpenAI has not announced a model by that name, and gpt-6.2-astra is not in OpenAI's API model list. What we have is Sam Altman saying the next GPT is coming soon, plus leaked names that OpenAI hasn't confirmed.
- Altman: the next GPT is coming soon. On stage at ADIPEC in Abu Dhabi on November 2, Altman said OpenAI had solved one of the Millennium Prize problems with 10,000 agents running an upcoming GPT model for about four days (The National). Attendees quoted him on X as calling that model the next GPT, to be released soon (AGTP). There is no official transcript, and he gave no name or date.
- What people saw in Codex: from November 1, several Codex users posted usage breakdowns that list
gpt-6.2-astranext togpt-6-astra, while the Codex model picker still says GPT-6 Astra (TokenPost, FourWeekMBA, TestingCatalog). Other users don't see it in their own records. - GPT-6.1 Astra is off the table: OpenAI shelved it on September 28, 2026 after it missed the company's safety and alignment bar (The Verge).
- The name isn't settled: a screenshot of an internal tracker, labeled "INTERNAL", shows
gpt-6.2-betelgeuse, which some link to a model reported as "Bel". Posts on X predicted a November 1 release; it didn't happen, and on November 2 Kingy AI found no GPT-6.2 entry in OpenAI's model directory, release notes or API changelog. Neither name is confirmed. - A reported date, not confirmed: on November 4, X user @kimmonismus, citing The Information, posted that GPT-6.2 could come as soon as Friday, November 6, in Astra-, Sol- and Cyber-focused versions (post). The Information's article is paywalled, and OpenAI hasn't confirmed any of this.
- What you can use now: GPT-6 Astra (
gpt-6-astra), OpenAI's most capable GPT-6 model, released September 3, 2026. It works with a TeamoRouter key in Codex, in Claude Code and over the API (model page).
New accounts start with a $0 balance — top up before your first request. Top up from $5. If a request returns 402 (400 on the OpenAI format) with "wallet balance is insufficient", your balance is empty — your base URL and key are fine. Top up and retry.
Before GPT-6.2 ships, use GPT-6 Astra with the same key
What the Codex usage screenshots show
- The label appears in Codex usage analytics, not in the model picker or the API. Users can't select
gpt-6.2-astrathemselves. - In every breakdown TestingCatalog saw,
gpt-6.2-astrashows up only on the October 31 bar. Its share varies by account: 1.6% of one user's week, 19.8% of another's, and 37.7% of a single day for a third. - That pattern looks like some Codex requests being routed to a new model as a test, but OpenAI has not said so, and the same handful of screenshots is behind every report.
- OpenAI's own model docs still list GPT-6 Astra, with no 6.2 model, as of November 3, 2026.
What Altman said at ADIPEC
Altman spoke at the ADIPEC opening ceremony on November 2, 2026. The National reported that he said OpenAI had solved a Millennium Prize problem, using 10,000 agents that ran an upcoming GPT model for about four days with "no hierarchy", and noted that the claim could not immediately be verified. According to attendees' posts on X, he described the internal model behind that result as the next GPT and said it would be released soon. These are accounts of a live talk, not an OpenAI announcement.
The result he referred to is OpenAI's September 8 post On the Navier–Stokes Millennium Prize Problem, which shared an AI-generated solution with a formal proof in Lean. OpenAI said then that it came from an internal model more capable than GPT-6 Astra, with about 10,000 agents working for roughly 88 hours (The Next Web). The proof hasn't been independently verified, and Axios reported a dispute over credit with NYU mathematician Tristan Buckmaster, who posted related work the same day.
So the solid part is narrow: OpenAI's CEO says a new GPT model is coming soon and that it was the model behind the Navier–Stokes run. Its name, date, price and API availability are still unknown. We are not repeating benchmark or price rumors.
Timeline
| Date (2026) | What happened |
|---|---|
| September 3 | OpenAI releases GPT-6 Astra (gpt-6-astra) |
| September 28 | OpenAI says it won't release GPT-6.1 Astra after safety and alignment tests |
| September 29 | At DevDay, OpenAI launches GPT-6.1 Sol instead |
| October 26 | OpenAI launches GPT-6.1 Terra |
| September 8 | OpenAI posts its AI-generated Navier–Stokes solution; Axios reports a dispute over credit |
| October 30–31 | A screenshot showing gpt-6.2-betelgeuse circulates on X, with talk of a November 1 release |
| October 31 | The only day gpt-6.2-astra appears in the shared Codex usage breakdowns |
| November 1 | The rumored release date passes without a release |
| November 1–2 | TokenPost, FourWeekMBA and TestingCatalog report the Codex screenshots; OpenAI has not commented |
| November 2 | Altman says at ADIPEC that the next GPT, the model behind the Navier–Stokes result, is coming soon |
| November 4 | The Information reports, via @kimmonismus on X, that GPT-6.2 could ship as soon as Friday in Astra, Sol and Cyber versions; OpenAI hasn't confirmed |
What isn't known yet
When the next GPT ships, under what name (GPT-6.2 Astra, Betelgeuse or something else) and model ID, at what price, with what context window, and whether it reaches the API on day one. We are not going to guess.
On release day we'll update this article with whether the model is available on TeamoRouter. You can also vote for it on the GPT-6.2 Astra upcoming page. The models live on TeamoRouter are always listed on the models page.
Use GPT-6 Astra today
From OpenAI's GPT-6 Astra model page: text and image input, a 1,050,000-token context window, 128,000 max output tokens, and reasoning.effort from low to max. OpenAI's price is $10 per million input tokens, $1.00 cached input and $50 output, with 2x input and 1.5x output for prompts over 272K input tokens. TeamoRouter's current price is on the pricing page.
Codex CLI. Put this in ~/.codex/config.toml and set your TeamoRouter key in OPENAI_API_KEY:
model_provider = "teamorouter"
model = "gpt-6-astra"
[model_providers.teamorouter]
name = "TeamoRouter"
base_url = "https://api.teamorouter.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"
Claude Code. Point it at TeamoRouter and pass the model name:
export ANTHROPIC_BASE_URL="https://api.teamorouter.com"
export ANTHROPIC_AUTH_TOKEN="sk-teamo-your-key"
claude --model gpt-6-astra
Claude Code shows a notice that gpt-6-astra isn't in its built-in model catalog and keeps the session within 200K tokens of context. That is expected; requests still go to GPT-6 Astra.
We sent both setups to the TeamoRouter gateway on November 2, 2026 with Codex CLI 0.166.0 and Claude Code. Codex calls /v1/responses; Claude Code calls /v1/messages in the Anthropic format. More setup detail: GPT-6 Astra in Codex, Cursor and Claude Code.
If GPT-6.2 Astra ships: if it becomes available on TeamoRouter, change only the model name in the configs above to the ID OpenAI publishes. The base URL and key stay the same. Step-by-step for release day, including an SDK fallback to GPT-6 Astra: Switching Codex, the OpenAI SDK and Claude Code to GPT-6.2.
Astra, Sol or Terra today
| Model | OpenAI's description | OpenAI price (input / output per 1M) |
|---|---|---|
GPT-6 Astra (gpt-6-astra) |
Most capable, for the most demanding work | $10 / $50 |
GPT-6.1 Sol (gpt-6.1-sol) |
Near-Astra performance for complex coding, at lower cost | $2 / $10 |
GPT-6.1 Terra (gpt-6.1-terra) |
Fast and efficient: visual reasoning, computer use, writing | $1 / $5 |
All three are on TeamoRouter with the same key. In Codex, start a session with codex -m gpt-6.1-sol to switch.
Paying for usage
No subscription: you top up a balance from $5 and pay only for the tokens you use, by card or with USDT, USDC or ETH from any exchange app or wallet (how to pay with USDT). The same key also works for Claude, Gemini and DeepSeek.
FAQ
Is GPT-6.2 Astra out?
No. As of November 3, 2026, OpenAI has not announced it and it isn't in the API. Sam Altman said on November 2 that the next GPT is coming soon, without a name or date; gpt-6.2-astra has only been seen as a label in some Codex usage breakdowns. On November 4 The Information reported a possible release as soon as Friday, November 6, in Astra, Sol and Cyber versions; OpenAI hasn't confirmed it.
What is GPT-6.2 Betelgeuse, or "Bel"?
gpt-6.2-betelgeuse is an identifier seen in a screenshot of an internal tracker shared on X, and some posts link it to a model reported as "Bel". The source of the screenshot hasn't been verified, the predicted November 1 release didn't happen, and OpenAI hasn't used either name.
Can I use GPT-6.2 Astra on TeamoRouter?
Not yet, because OpenAI hasn't released it. GPT-6 Astra (gpt-6-astra), GPT-6.1 Sol and GPT-6.1 Terra are available now.
What happened to GPT-6.1 Astra?
OpenAI shelved it on September 28, 2026 after it fell short of the company's safety and alignment bar, and launched GPT-6.1 Sol the next day. There is no GPT-6.1 Astra model to use.
Why do I see gpt-6.2-astra in my Codex usage?
Codex chose which model served some of your requests; you didn't select it, and OpenAI hasn't explained the label. Your configured model is still the one in the picker.