- Latest News about Uncensored AI
- What Is GPT-6.2 Astra? OpenAI’s New Frontier Model and GPT-6.2 Sol (2026)
What Is GPT-6.2 Astra? OpenAI’s New Frontier Model and GPT-6.2 Sol (2026)
Short answer: GPT-6.2 Astra is OpenAI’s new frontier model for hard professional work, and GPT-6.2 Sol is the cheaper, everyday counterpart in the same release. OpenAI published both on 17 November 2026. They are available in ChatGPT Work and Codex on supported paid plans, and in the API as gpt-6.2-astra and gpt-6.2-sol. Official API docs give both a 1,050,000-token context window, 922,000 max input tokens, and 128,000 max output tokens. They take text and images and reply with text. They are not marketed as uncensored models — see Can GPT-6.2 Astra do NSFW?.
This page sticks to confirmed official specs and availability. Benchmark figures below are OpenAI’s own, with OpenAI’s footnotes. We did not invent scores or run our own lab tests.
TL;DR
| Item | Official fact (as of 18 November 2026) |
|---|---|
| What launched | GPT-6.2 Astra (frontier) and GPT-6.2 Sol (everyday workhorse) |
| Announcement | Introducing GPT-6.2 Astra and Sol, 17 November 2026 |
| Where to use them in ChatGPT | Open Work (and Codex). ChatGPT Learn is explicit about that path |
| Plans | Launch post: Plus, Pro, Enterprise, and Edu. Learn also lists Business. Enterprise/Edu: a workspace admin must enable the models |
| API IDs | gpt-6.2-astra, gpt-6.2-sol |
| Context / output | 1,050,000 context; 922,000 max input; 128,000 max output |
| Standard API price (per 1M tokens) | Astra $8 / $0.40 cached / $40. Sol $2 / $0.10 cached / $10 (same as GPT-6.1 Sol) |
| Safety labels | Both: Preparedness Critical (cybersecurity) and High (biological and chemical). Astra also High for AI Self-Improvement; Sol is below that line |
Astra vs Sol
OpenAI positions the pair as two products, not one model with a nickname.
GPT-6.2 Astra is the flagship for the hardest projects: long professional documents, slides and spreadsheets that follow a template, software work, computer use, and scientific workflows. The launch post says its office files “read less like AI,” and that it writes cleaner code and stronger frontends, especially for websites, games, and 3D work. Official API docs describe it as the most capable model for work “from start to finish,” with text and image input and text output.
GPT-6.2 Sol is the default everyday model. Compared with GPT-6.1 Sol, OpenAI says 6.2 Sol is smarter, more efficient, and lets you finish more tasks on the same paid plan. ChatGPT Learn’s examples are practical: review a pull request, revise code after feedback, turn meeting notes into a brief.
A simple split:
| Choose | When |
|---|---|
| Astra | A hard, long-horizon job — a research question, a messy bug across services, a deck that has to match a house style |
| Sol | Everyday coding, writing, and multi-step work where cost and volume matter |
Both reject API requests that set reasoning effort to none. Official model cards list the same context and output limits. Their knowledge cutoffs differ: Astra is 18 May 2026; Sol is 1 June 2026.
Availability and plans
The launch post says that, starting on announcement day, both models are available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Eligible Enterprise and Edu users get that access once a workspace admin enables it. ChatGPT Learn repeats the Work/Codex path, also lists Business, and says: to use GPT-6.2 Astra or Sol in ChatGPT, open Work.
On supported consumer plans, new Work chats default to Sol at Light reasoning effort. Enterprise workspaces with model access default to Sol at Medium. If you see the Power slider, Learn says to open Advanced to pick a model and reasoning effort.
Ultrafast is plan-specific. The launch post: Pro 500 subscribers can use Ultrafast with either model; this is the first time Ultrafast is offered to Pro 200 subscribers, starting with Sol. Learn uses the same split as “Pro $200” and “Pro $500,” and notes that Codex and Work share usage.
Developers get both IDs in the OpenAI API, with Ultrafast available for both at launch. Learn adds a data-residency note: both models support US and EU regions for Standard, Fast, and Batch; Ultrafast supports US data residency only.
This page does not claim the models appear in every ChatGPT picker. Official docs route you to Work and Codex.
Official specs
From the GPT-6.2 Astra and GPT-6.2 Sol API model cards, unless noted:
| Spec | GPT-6.2 Astra | GPT-6.2 Sol |
|---|---|---|
| Model ID | gpt-6.2-astra | gpt-6.2-sol |
| Context window | 1,050,000 tokens | 1,050,000 tokens |
| Max input | 922,000 tokens | 922,000 tokens |
| Max output | 128,000 tokens | 128,000 tokens |
| Input | Text and image | Text and image |
| Output | Text | Text |
| Not supported as model modalities | Audio, video | Audio, video |
| Knowledge cutoff | 18 May 2026 | 1 June 2026 |
Reasoning effort none | Rejected | Rejected |
| Chat Completions + Responses | Supported (changelog) | Supported |
| Tool calling | Changelog: requires the Responses API | Same |
They are language/reasoning models with vision understanding, not OpenAI’s image-generation model line. The image-generation HTTP endpoints are listed as unsupported on these cards.
Pricing (API)
The launch post and the model cards agree on Standard rates per million tokens:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Astra | $8 | $0.40 | $40 |
| GPT-6.2 Sol | $2 | $0.10 | $10 |
Astra’s input and output prices are 20% lower than GPT-6 Astra, with a 60% cached-input discount. Sol keeps GPT-6.1 Sol’s token prices; OpenAI’s pitch is more work per dollar from better quality and fewer output tokens, not a sticker-price cut.
The official model cards add important billing footnotes, which the launch post does not spell out in the same place:
- For requests with more than 272K input tokens, input and cache rates double, and output costs 1.5× the short-context rate, for the full request.
- Batch is 50% of Standard. Fast is 2× Standard for the same context length.
- Ultrafast is 3× Standard through 28 February 2027, then 6× from 1 March 2027 (Pacific time).
- Cache-write prices and 30-minute vs 12-hour retention are on the model cards and the pricing page. This article does not flatten every tier into one number.
ChatGPT Work/Codex usage still depends on your subscription and shared usage pool. That is not the same meter as raw API token prices.
What OpenAI claims on benchmarks
These are OpenAI’s published comparisons, not our measurements. The launch post includes footnotes; two matter:
- Agents’ Last Exam results for Claude Fable 5.2 exclude nine tasks that used a fallback model.
- Cybersecurity charts in the post use internal configurations with some safeguards disabled; full methods are in the system card.
OpenAI says:
- On Agents’ Last Exam (long-horizon professional work), GPT-6.2 Astra scores 5.2 points higher than GPT-6 Astra at less than one-third the cost, and 5.5 points ahead of Claude Fable 5.2 at 5% of the cost.
- On Terminal-Bench-Science 0.1, Astra scores 9 percentage points higher than GPT-6 Astra at less than one-third the cost. OpenAI also claims a new high on the Vals Science Index.
- On Cognition’s FrontierCode 2.0, GPT-6.2 Sol comes within 0.5 points of GPT-6.2 Astra at about one-fifth the cost (both at max).
- On OpenAI’s LifeVal everyday-task eval, Sol outscores GPT-6 Astra at about one-seventh the cost.
Treat those as vendor results. Third-party league tables exist; this page does not promote them as official specs.
Safety, in brief
Both models, like GPT-6 Astra, meet the Preparedness Framework Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities. GPT-6.2 Astra is also OpenAI’s first externally released model to meet High for AI Self-Improvement; Sol does not. The launch includes tighter default limits on higher-risk cybersecurity help. The system card describes an early jailbreak-resistance regression versus GPT-6 Astra and a later adjustment of refusal boundaries.
That is policy and official evaluation language, not a green light for adult or “uncensored” use. For filters, host wording, and alternatives, read Can GPT-6.2 Astra do NSFW?.
How this differs from GPT-6 and GPT-6.1
GPT-6.2 is a new pair, not a rename of GPT-6 Astra or of GPT-6.1 Sol vs Terra. Official comparisons are against those earlier IDs. If you still have prompts, evals, or price sheets built for gpt-6-astra or gpt-6.1-sol, check the GPT-6.2 migration notes and the API changelog before you assume drop-in behavior.
FAQ
What is GPT-6.2 Astra?
OpenAI’s 17 November 2026 frontier model for difficult professional, coding, and scientific work. API ID: gpt-6.2-astra.
What is GPT-6.2 Sol?
The efficient sibling for everyday work and agents. API ID: gpt-6.2-sol. Same context and output limits as Astra; a later knowledge cutoff (1 June 2026) and a lower Standard token price.
Can I use GPT-6.2 in regular ChatGPT chat?
Official Learn docs say to open Work. This page does not claim a regular Chat picker entry that those docs do not describe.
Is GPT-6.2 uncensored?
No official page markets it that way. See Can GPT-6.2 Astra do NSFW?.
Where should I go for private adult chat instead?
HackAIGC Chat is built for legal adult use. That is a product path, not a claim that it is GPT-6.2.
Related reading
- Can GPT-6.2 Astra Do NSFW?
- Kling Image 4.0 review (a new image model from the same news window; not an OpenAI product)
- What Is GPT-6.1? Sol vs Terra
- Can GPT-6.1 Do NSFW?
- Can GPT-6 Astra generate NSFW?
Sources: OpenAI launch post; ChatGPT Learn: GPT-6.2; API model cards for Astra and Sol; API changelog; GPT-6.2 system card.
