Checked November 11, 2026 — still no GPT-6.2
OpenAI's changelog ends at gpt-6-cyber (Nov 5), and the models page lists only gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna and gpt-6-cyber for the current family. The model docs for gpt-6.2-astra and gpt-6.2-sol return 404. If that changes, this page is updated on the same URL.
Compare the confirmed GPT-6 Astra with Claude Fable 5.1Rumor tracker · checked November 11, 2026
GPT-6.2 and GPT-6.2 Sol: confirmed vs rumor
OpenAI's next Astra/Sol step has not shipped. Here is the dated evidence trail, the fields that are still unpublished, and how the confirmed GPT-6 family compares with Claude Fable 5.1 today.
TL;DR
GPT-6.2 is not released. As of November 11, 2026 OpenAI's API changelog still ends at GPT-6 Cyber (November 5), the model catalog lists no 6.2 ID, and the gpt-6.2-astra and gpt-6.2-sol documentation pages both return 404. Everything circulating — a Codex usage label, Google Cloud quota screenshots, a "Bel"/Betelgeuse codename, users saying they are routed to "6.2 Sol" — is unconfirmed reporting. No price, context window, modality or benchmark for either model has been published, so none is quoted here.
Timeline: how the GPT-6.2 story got here
Official releases and the reports around them, in the order they happened. Every row is labelled with the kind of evidence behind it.
| Date | What happened | Evidence |
|---|---|---|
| Sep 3, 2026 | GPT-6 Astra (gpt-6-astra) released as OpenAI's flagship. | Official — OpenAI changelog |
| Sep 22, 2026 | GPT-6 Sol and GPT-6 Luna released at $2/$10 and $0.10/$0.50 per M tokens. | Official — OpenAI changelog |
| Sep 29, 2026 | GPT-6.1 Sol (gpt-6.1-sol) released at the same published $2/$10 for prompts up to 272K input tokens. | Official — OpenAI changelog |
| Oct 26, 2026 | GPT-6.1 Terra released, aimed at visual reasoning and computer use. | Official — OpenAI changelog |
| Oct 29–30, 2026 | RuntimeWire, citing sources, identifies an internal "Bel" as GPT-6.2 Betelgeuse, built on a "Dougtrio" pre-train series. | Reported — single outlet, no model card |
| Nov 1, 2026 | Developers post screenshots of Codex usage analytics that contain a gpt-6.2-astra label. | Reported — screenshots only |
| Nov 4, 2026 | The Information reports new Sol, Astra and cyber models "could be coming as soon as Friday"; X users post Google Cloud quota screenshots listing gpt-6.2-astra and gpt-6.2-sol. | Reported — press, no OpenAI confirmation |
| Nov 5, 2026 | GPT-6 Cyber ships — the only one of the three reported models that arrived. | Official — OpenAI changelog |
| Nov 8, 2026 | ChatGPT and Codex users say they are being routed to "GPT-6.2 Sol", citing stronger single-file Three.js output and longer Plus limits; no post shows an API model ID. | Reported — user posts, no model ID |
| Nov 11, 2026 | This check: changelog ends at Nov 5, the catalog has no 6.2 ID, both 6.2 model-doc URLs 404. | Verified — OpenAI pages, Nov 11 |
What is known about GPT-6.2 and GPT-6.2 Sol
Short answer: the model identifiers are not published, so almost nothing downstream of them is either.
| Item | Status as of November 11, 2026 |
|---|---|
| Model ID | Not published — gpt-6.2-astra and gpt-6.2-sol doc pages return 404. |
| Price | Not published — no 6.2 row on OpenAI's pricing page, in any tier. |
| Context window and max output | Not published — no window or max-output figure. |
| Modalities | Not published — the Astra/Sol lineage implies text and image input, but 6.2 itself is undocumented. |
| Benchmarks and size | Not published — the "over 10 trillion parameters" figure traces to an unverified August post. |
| Release date | No date from OpenAI; a reported November 6 window passed with only GPT-6 Cyber. |
| API and cloud availability | Not in OpenAI's model catalog, not on OpenRouter. |
| Codename | "Bel" / Betelgeuse is a single-outlet report, not an OpenAI name. |
Anything beyond this table — a rate, a window, a benchmark percentage — would be a guess. This page does not print guesses; the numbers it does show come from the model registry.
How the rumored GPT-6.2 compares with Claude Fable 5.1
Only the confirmed models have numbers. Rates and windows below come from this site's model registry, verified against each vendor's own pricing page.
| Spec | GPT-6.2 (rumored) | GPT-6 Astra | GPT-6 Sol | Claude Fable 5.1 |
|---|---|---|---|---|
| Model ID | Not published | gpt-6-astra | gpt-6-sol | claude-fable-5-1 |
| Input (per M tokens) | Not published | $10.00 | $2.00 | $10.00 |
| Output (per M tokens) | Not published | $50.00 | $10.00 | $50.00 |
| Cached input (per M tokens) | Not published | $1.00 | $0.20 | $0.25 |
| Long-context billing | Not published | 2x input / 1.5x output above 272K input | 2x input / 1.5x output above 272K input | Flat rates at any context length |
| Context window | Not published | 1.05M | 1.05M | 1M |
| Max output | Not published | 128K | 128K | 128K |
| Data retention | Not published | Zero Data Retention for eligible API customers | Zero Data Retention for eligible API customers | 30-day mandatory |
The GPT-6.2 column is empty on purpose: OpenAI has published no rate, window or spec, so there is nothing to compare yet. The other columns are registry rows last verified against vendor pages on 2026-09-25 (Astra), 2026-10-11 (Sol) and 2026-09-14 (Fable 5.1).
What to do while you wait
A rumored dot release is not a reason to move production traffic.
Stay on documented IDs
gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra and the task-specific gpt-6-cyber are the IDs OpenAI documents today. Nothing named 6.2 can be called, so there is no config change to make yet.
Wait for three receipts
A changelog row, a model page that is not a 404 with a published rate, and the ID appearing in your own /v1/models response. Usage labels and quota screenshots have never been a launch notice.
Optimize the harness instead
GPT-6.1 Sol already lists at a fifth of GPT-6 Astra's token price (September 29 changelog). Prompt caching, reasoning effort and context discipline move cost further than a dot release will.
Pilot against Fable 5.1
If what you want from 6.2 is agentic coding, Claude Fable 5.1 already publishes an 87.7% SWE-bench Verified score. When a 6.2 model ships with a system card, test it on your own frozen tasks, not on a demo clip.
This page tracks the rumor; it does not price it. When OpenAI publishes a 6.2 rate card, the row goes into the registry and the table above fills in.
Sources checked on November 11, 2026
Primary pages first, then the reporting the rumor rests on.
- OpenAI API changelog
Newest entry is gpt-6-cyber, November 5, 2026. No 6.2 row.
- OpenAI model catalog
Lists gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna, gpt-6-cyber.
- OpenAI API pricing
Standard, batch and ultrafast tables list no 6.2 model.
- Model doc for gpt-6.2-sol
Returns 404, as does the page for gpt-6.2-astra.
- RuntimeWire, October 29, 2026
The report that identifies "Bel" as GPT-6.2 Betelgeuse.
Compare the confirmed models now
The GPT-6 Astra comparison runs the full spec and benchmark table against Claude Fable 5.1, using both vendors' published rates.