Is GPT-6.2 Out? What's Confirmed, What's Still a Leak, and What You Can Call Today

A usage-table label is not a product name. On the evening of 12 November 2026, Asia/Shanghai, OpenAI has not officially released or announced GPT-6.2, and SandBase does not list it.

The API changelog, the models catalog, the pricing table, the company newsroom, and the product release notes were all checked that evening. None of them has a GPT-6.2 entry. Candidate SandBase paths returned HTTP 404 on both the HTML card and the markdown file. llms-full.txt and the sitemap, both HTTP 200, contain zero gpt-6.2 or gpt-6-2 hits.

The strings you can send are the ones OpenAI already ships: GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, and GPT-6 Luna. GPT-6 Sol is the previous Sol, still in the catalog. GPT-6-Cyber, dated 5 November, is a trusted-access cyber model. It is not a 6.2 launch, and it is not on SandBase.

Key takeaway

• Not officially released, and not on SandBase, as of 12 November 2026, Asia/Shanghai. There is no official product name “GPT-6.2” and no official API ID.
• Docs for gpt-6.2, gpt-6.2-astra, gpt-6.2-sol, and gpt-6.2-terra returned 404. Those strings are not shipping SKUs.
• Official shipping names: GPT-6 Astra (gpt-6-astra), GPT-6.1 Sol (gpt-6.1-sol), GPT-6.1 Terra (gpt-6.1-terra), GPT-6 Luna (gpt-6-luna). GPT-6 Sol (gpt-6-sol) is the older Sol.
• November entries so far: ChatGPT Images 2.6 on 4 Nov, GPT-6-Cyber for authorized vulnerability research on 5 Nov, Business Hub in ChatGPT on 10 Nov. No 6.2 note.
• On SandBase today: GPT-6 Astra at $9 / $45 per 1M (10% off official $10 / $50), GPT-6.1 Sol at $2 / $10, GPT-6.1 Terra at $1 / $5, GPT-6 Luna at $0.10 / $0.50. Route: POST https://api.sandbase.ai/v1/chat/completions.

Short answer: not officially, and not on SandBase

The leak posts and the catalog do not say the same thing. The catalog is the one a request has to match. I am dating the verdict to the evening of 12 November 2026, Asia/Shanghai, and I am not going to put a 6.2 string in a client on the back of a screenshot.

This environment could not screenshot developers.openai.com. The changelog, models catalog, and pricing page returned HTTP 403 from here, so this article has no image of those three pages. They were read as official markdown and are linked above. The crop in the November section is the product release notes, which did load. It is not a picture of the changelog.

The same catalog check, run the other way, is the Gemini 4 Carbon status note from 11–12 Nov. Carbon has a public ID and a SandBase card, which is why the routing note and the try-today note exist. There is no card like that for a 6.2 ID.

Officially confirmed: the GPT-6 family you can name today

Official names and API IDs

Product names use the vendor spelling: a space, and a point in “GPT-6.1”, not a hyphen. “GPT-6.1 Sol” is the product. gpt-6.1-sol is the ID. “GPT-6.1-Sol” and “GPT-6-Astra” are misspellings of the product name.

The catalog featured list is GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, and GPT-6 Luna. Browse also includes GPT-6 Sol and GPT-6-Cyber. It does not include GPT-6.2. The catalog’s starting line, fetched the same evening:

If you’re not sure where to start, use GPT-6 Astra, our flagship model for complex reasoning and coding. Choose GPT-6.1 Sol to balance intelligence and cost, or GPT-6 Luna for cost-sensitive, high-volume workloads.

The Using GPT-6 guide says to set model to gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, or gpt-6-luna, and to pick Astra for the most demanding work, 6.1 Sol for complex work at lower cost, 6.1 Terra for visual reasoning and writing, or Luna for focused high-volume tasks. That is the official four.

Official context on these general models is a 1,050,000-token window, 922,000 max input, and 128,000 max output. SandBase cards print 1.1M context and 128K max output. Cite 1,050,000 for OpenAI. Cite 1.1M for a SandBase card. Do not “correct” the card in the same sentence.

The guide mentions pro mode, carried over from GPT-5.6. Official docs for gpt-6-astra-pro returned 404. SandBase IDs such as openai/gpt-6-astra-pro (and the Sol, Terra, Luna, and older-Sol pro IDs) are catalog routes for pro mode of the same model, not official model-page IDs and not different weights. Pro mode can spend more tokens, so unit price is not task price.

Two gated names sit next to that family. Neither is a 6.2 alias.

GPT-6-Cyber (gpt-6-cyber) is the 5 November changelog entry. Access is Daybreak Red / approved organizations, for authorized vulnerability research. Official Standard short context is $20 / $100 / $2 cached / $25 for a 30-minute cache write, per 1M tokens. Long context on that row is $40 / $150 / $4 cached input. It is not on SandBase, so there is no SandBase price. This page does not describe an exploit or how to reproduce a vulnerability. The 5 Nov ship is not the next Astra or the next Sol.

GPT-6 Rosalind is the life-sciences trusted-access SKU from the 30 October changelog. The changelog says GPT-6 Rosalind / gpt-6-rosalind. The catalog display is GPT-Rosalind / gpt-rosalind-research. Pricing lists both. Both model pages 404, so this note does not pick one canonical ID. Short-context list on that row is $15 / $75 / $1.50 / $18.75 per 1M. Not on SandBase.

Official Standard prices

Official Standard, short context (272K input tokens or fewer), from the pricing table and the model pages. USD per 1M tokens, fetched 12 Nov 2026.

Cached input on GPT-6.1 Sol is 5% of uncached input ($0.10). On the older GPT-6 Sol it is 10% ($0.20). Same $2 / $10 sticker, different cache bill. New work should use GPT-6.1 Sol unless a client is already pinned to gpt-6-sol.

Long context (more than 272K input tokens) is 2× input and cache, and 1.5× output, for the full request. Batch and Flex are 50% of Standard. Fast is 2×. Ultrafast, listed for gpt-6-astra and gpt-6.1-sol, is 6×. Regional processing and FedRAMP add 10% for models released on or after 5 March 2026, which includes this family. That uplift is an official-endpoint note, not a SandBase line. The official table also has a longer cache-write tier. The column above is the 30-minute write, so it can sit next to the SandBase heroes.

Introducing GPT-6.1 Terra is the newest general GPT-6 announcement: Plus, Pro, Enterprise, and Edu in ChatGPT Work and Codex, API ID gpt-6.1-terra, standard API $1 / $0.05 cached / $5 per 1M. The newsroom’s latest product-model post is still that 26 October piece. A 7 October item, “GPT-6 and Intelligent UI for everyone,” is older, and its body was not re-quoted here.

Arithmetic only, not a measured job. 100,000 uncached input tokens and 10,000 output tokens: Astra $1.50, GPT-6.1 Sol $0.30, GPT-6.1 Terra $0.15, Luna $0.015, all at official short-context list. If a real turn thinks far past 10,000 output tokens, this understates the bill. Ultrafast at 6× turns the Astra example into $9.00 before the prompt changes.

What shipped in November so far

• 2 Nov. Outbound SIP in the Live API. Not a GPT text model.
• 4 Nov. ChatGPT Images 2.6 and GPT Image 2.6 Sunburst and Flare. Release notes also mention Dots for Enterprise, Edu, and Healthcare that day.
• 5 Nov. GPT-6-Cyber for authorized vulnerability research. Latest November model entry in the changelog. Not GPT-6.2.
• 10 Nov. Business Hub in ChatGPT.

Other November newsroom items that evening (veterans’ ChatGPT Plus, education and freelance access, Australian universities, the Pierre Computer acquisition) are not model launches.

Official OpenAI product release notes as of 12 Nov 2026, Asia/Shanghai. Visible November entries include Business Hub (10 Nov), GPT-6-Cyber for authorized vulnerability research (5 Nov), and ChatGPT Images 2.6 (4 Nov). There is no GPT-6.2 note.

The crop is the release-notes page only. It does not show the API changelog, the models catalog, or the pricing table. The Cyber row is badged GA on that page, and so are Business Hub and Images 2.6, so the badge is the release-notes status chip. It is not a statement that every API key can call gpt-6-cyber. The model page still limits that ID to Daybreak Red / approved organizations. This note does not repeat the page’s line about reproducing a vulnerability. Those three stayed at HTTP 403 for a screenshot, and this note does not invent a picture of them. The changelog text, read the same evening, matches the list: latest November model entry is GPT-6-Cyber on 5 Nov, and there is no GPT-6.2 line.

What’s still a leak (and what those leaks actually show)

None of the following is a source for an official name, an ID, or a price.

The Codex usage label gpt-6.2-astra

Reported. CellCog (around 1 Nov 2026), TestingCatalog (2 Nov 2026), and ArtificialWatch describe X screenshots from the morning of 1 November 2026 UTC: ChatGPT or Codex usage analytics with a line labelled gpt-6.2-astra beside gpt-6-astra, often on a 31 October bar. One commonly cited split is 80% / 19.8% on a single account. That is one account, not a fleet statistic.

CellCog and ArtificialWatch attribute a cropped response.completed event, "model": "gpt-6.2-astra", to @Mr_Salio at 10:30 UTC on 1 Nov. This turn did not fetch those original posts or open the images, so there is no pixel caption here. Users posted screenshots that appear to show the label. Official docs for gpt-6.2-astra returned 404. The catalog does not list it. Do not send the string.

Google Cloud quota strings

Techbooky describes a screenshot @AiBattle_ posted on 4 November 2026: a Cloud Console quotas filter with base_model:gpt-6.2-astra and base_model:gpt-6.2-sol. Techbooky says this is not confirmation of customer availability. The official Astra page lists Google Cloud private preview for gpt-6-astra only. A filter string is not “listed on Google Cloud.”

The early-November press window

The Information’s AI Agenda page returned HTTP 403, so this note does not quote it and does not say The Information announced GPT-6.2. Secondaries fetched at HTTP 200, Orca Router, 4SAPI (dated 5 Nov 2026), and CellCog, paraphrase a 4 November item: a next-generation Astra, a new Sol, and cyber-focused models as soon as Friday 6 November, timelines could change. They agree the report did not print “6.2”, an ID, or a price. 6 November passed with no official 6.2 launch and no new general Astra or Sol. GPT-6-Cyber on 5 November is the trusted-access model, not that rumor.

A ChatGPT-routing rumor and a Bel / gpt-6.2-betelgeuse string were not checked against a primary. They are not evidence. No verified official sentence says a 6.2 model is already in the API.

Tibo’s 11 Nov post is a banked-reset hint, not a model name

Tibo (@thsottiaux), bio “Codex & ChatGPT @OpenAI,” posted at 2026-11-11T06:57:08Z: 11 November 2026, 14:57:08 Asia/Shanghai.

Thursday is feeling codexy to me. Perhaps also a good time for a banked reset.

The parent in the same thread, status/2120410006273380750, says: “It’s been a while without resets. Maybe I should come back out of the basement.” Neither post contains “GPT-6.2”. Thursday from a Wednesday 11 Nov post is 12 November 2026. A banked reset is a redeemable usage-limit coupon, not a model name. Whether one landed on 12 Nov is unknown. This is not a launch teaser.

What you can actually call today on SandBase

Live cards and honest prices

Rechecked 12 November 2026, 23:17–23:18 Asia/Shanghai. HTML and markdown both 200. The OpenAI index lists these cards and does not list a 6.2 model. Cache columns are the HTML hero. Input and output match the card markdown.

SandBase does not bill official list on Astra or on GPT-6 Sol. Those cards show 10% off. GPT-6.1 Sol and GPT-6.1 Terra match list. Do not copy Astra’s discount onto them.

Live SandBase GPT-6 Astra card at /models/openai/gpt-6-astra, captured 12 Nov 2026, Asia/Shanghai (HTTP 200). Hero is $9.00 / $45.00 (10% off the $10 / $50 list), with cache read $0.90 and cache write $11.25 per 1M tokens.

That 10% is the card’s short-context gap ($1.35 vs $1.50 on the 100,000-in / 10,000-out example). It is not a quote for long context, Fast, or Ultrafast.

Live SandBase GPT-6.1 Sol card at /models/openai/gpt-6.1-sol, captured 12 Nov 2026, Asia/Shanghai (HTTP 200). Hero is $2.00 / $10.00 / cache read $0.10 / cache write $2.50 per 1M tokens, matching official Standard short-context list, with no discount shown.

GPT-6.1 Terra is the card for visual reasoning, computer use, and writing when you want the newest general GPT-6 and you do not need Astra. $1 / $5 on the list and on the card. GPT-6 Luna is the volume option at $0.10 / $0.50, hero cache write $0.13. The cards say 1.1M / 128K. Official pages say 1,050,000 / 128,000. Budget against the card you are about to call.

Pro-mode cards were live the same evening, with no discount badge. openai/gpt-6-astra-pro is $10 / $50 / $1.00 / $12.50, official list, not the 10% on standard Astra. openai/gpt-6.1-sol-pro matches 6.1 Sol. openai/gpt-6.1-terra-pro matches Terra. openai/gpt-6-luna-pro is $0.10 / $0.50 with cache write $0.13. openai/gpt-6-sol-pro is list $2 / $10 / $0.20 / $2.50, not the 10% on standard GPT-6 Sol. Same unit price can still cost more if pro mode spends more tokens.

Already-written routing, all about models you can call: Fable 5.2 vs Opus 5.5 vs GPT-6 Astra (9–10 Nov), Claude Sonnet 5.5 vs GPT-6.1 Sol, and GPT-6 Astra usage limits and API cost. High-volume Luna notes: the public-opinion agent and the 22-model tier benchmark. September’s Astra multimodal checks are about GPT-6 Astra, not a later point release.

How to call them

Live cards document POST https://api.sandbase.ai/v1/chat/completions and POST https://api.sandbase.ai/v1/responses. The first API call uses chat completions. /v1/run is the media route. Do not send these models there.

No paid run was required for this note. The example targets a live ID, openai/gpt-6.1-sol, because that card matches official list. Swap in openai/gpt-6-astra, openai/gpt-6.1-terra, or openai/gpt-6-luna when that is the card you mean. There is no 6.2 model string in the body.

curl -X POST https://api.sandbase.ai/v1/chat/completions \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "openai/gpt-6.1-sol",
  "messages": [
    {"role": "user", "content": "Hello"}
  ],
  "max_tokens": 2048
}'

Sign in is where /signup redirects. Create a key, then send the call above.

What we will check if GPT-6.2 is ever announced

This page stays a status note until all four of these exist:

1. A changelog entry that names the model and an API ID.
2. A model page that returns HTTP 200, not a 404 for gpt-6.2, gpt-6.2-astra, gpt-6.2-sol, or gpt-6.2-terra.
3. A pricing-table row for that ID.
4. After that, a SandBase HTML card and the matching .md, both HTTP 200. A 404 is not a URL to publish.

Until then there is no catalog link here for a 6.2 model. If OpenAI ships the next step under a name it already uses, the update belongs on the Astra, Sol, Terra, or Luna card.

FAQ

Is GPT-6.2 out?

No. As of 12 November 2026, Asia/Shanghai, it is not in the changelog, catalog, pricing table, newsroom, or release notes, and not on SandBase. A reported usage label is not a model the docs let you select.

What is the API model ID?

There is no official one. gpt-6.2, gpt-6.2-astra, gpt-6.2-sol, and gpt-6.2-terra 404 as docs. Name gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna, or older gpt-6-sol, and prefix openai/ on SandBase. gpt-6-cyber is official, trusted-access only, and not a 6.2 ID.

What can I call on SandBase today?

GPT-6 Astra at $9 / $45 (10% off official $10 / $50), GPT-6.1 Sol at $2 / $10, and GPT-6.1 Terra at $1 / $5. GPT-6 Luna is $0.10 / $0.50, hero cache write $0.13. Links are in the table above. Endpoint: POST https://api.sandbase.ai/v1/chat/completions.

Did Tibo’s 11 November post announce a model?

No. The two sentences are a “codexy” Thursday and a possible banked reset. “GPT-6.2” is not in the post. A banked reset is a usage coupon. Thursday was 12 November 2026.

Is GPT-6-Cyber the November GPT-6.2 release?

No. gpt-6-cyber is 5 November 2026, Daybreak Red / approved organizations, for authorized vulnerability research. It is not on SandBase, and the docs do not call it GPT-6.2.

Sources

Confirmed on 12 November 2026, Asia/Shanghai: changelog, catalog, pricing, Using GPT-6, model pages for Astra, GPT-6.1 Sol, GPT-6.1 Terra, Luna, GPT-6 Sol, and GPT-6-Cyber, the newsroom, release notes, and Introducing GPT-6.1 Terra. Docs for gpt-6.2, gpt-6.2-astra, gpt-6.2-sol, gpt-6.2-terra, and gpt-6-astra-pro returned 404. Live SandBase cards above returned 200 on HTML and markdown between 23:17 and 23:18. Candidate 6.2 paths returned 404 on both.

Reported only, not used for names, IDs, or prices: CellCog, TestingCatalog, and ArtificialWatch on the usage label; Techbooky on the quota filter; Orca Router, 4SAPI, and CellCog on the press window whose primary was not fetched. The Tibo posts were fetched. They do not name GPT-6.2.

Open GPT-6 Astra, GPT-6.1 Sol, or GPT-6.1 Terra for a model that exists tonight. Luna stays the volume option when the token count is the bill.

Sources checked 12 November 2026, Asia/Shanghai. Official pages for names, IDs, and prices. Third-party pages only inside the leak section. No screenshot of the developer changelog, catalog, or pricing page: those hosts returned HTTP 403 from this environment.