OpenAI launched two new models on Tuesday 17 November 2026: GPT-6.2 Astra, its frontier model for the hardest professional work, and GPT-6.2 Sol, an efficient model for everyday tasks. Astra costs 20% less than the model it replaces, Sol keeps the same price as before, and both are available from launch day in ChatGPT Work, Codex and the API. There is one catch worth knowing before you go looking for them: neither model is in regular ChatGPT Chat.
What OpenAI actually launched
GPT-6.2 Astra is the new flagship, built for demanding professional work such as financial analysis, legal research and complex coding. GPT-6.2 Sol is the everyday model, for quick coding, writing and multi-step tasks — OpenAI says many of its own employees now use Sol as their go-to. The split is deliberate: one model for the jobs where accuracy matters most, one for the jobs where speed and price matter more.
Both models launched on 17 November 2026 in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, and in the API as gpt-6.2-astra and gpt-6.2-sol. An OpenAI staff reply on the company’s developer forum confirms the rollout does not add either model to regular ChatGPT Chat, and that availability depends on plan, workspace permissions and rollout progress. If you open ChatGPT and cannot find them, that is why.
On paper the two models share the same shape: a 1.05-million-token context window, up to 128,000 output tokens per response, and a knowledge cutoff in mid-2026 — 18 May for Astra, 1 June for Sol. What differs is what they are tuned for, and what they cost.
What GPT-6.2 Astra and Sol cost
Astra is the model that got cheaper. Sol is the model that stayed the same price — and that is the point. Astra’s standard API rates fall 20% against GPT-6 Astra, and cached input falls 60%. Sol keeps GPT-6.1 Sol’s prices exactly, which makes it a quarter of Astra’s price on input and output.
| Model | Input (per 1M tokens) | Cached input | Output (per 1M tokens) |
|---|---|---|---|
| GPT-6.2 Astra | $8.00 | $0.40 | $40.00 |
| GPT-6 Astra (previous) | $10.00 | $1.00 | $50.00 |
| GPT-6.2 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6.1 Sol (previous) | $2.00 | $0.10 | $10.00 |
Two pieces of fine print matter. First, these headline rates apply to short-context requests of up to 272,000 input tokens; above that, input and cache rates double and output costs 1.5 times as much for the whole request. Second, OpenAI cut Ultrafast pricing by 50% for three months, and Pro subscribers at $200 a month get Ultrafast for the first time, starting with Sol — Pro subscribers at $500 a month get it on both models.
How they score: OpenAI’s numbers vs independent tests
OpenAI’s own benchmarks say Astra is a clear step up. Independent testing agrees on the direction, but not on the size of the saving. That gap is the most useful thing in this launch, because it tells you how much of the improvement is the model and how much is the price cut.
On OpenAI’s own evaluations, Astra scores 5.5 points ahead of Anthropic’s Claude Fable 5.2 on Agents’ Last Exam — a benchmark for multi-step professional workflows — at 5% of the cost, and Sol comes within 0.5 points of Astra on Cognition’s FrontierCode 2.0 at about a fifth of the cost. Astra also leads the AA-3D benchmark at 1,318 Elo, ahead of Fable 5.2 at 1,271. These are OpenAI’s chosen test sets, so treat them as the vendor’s case, not a neutral verdict.
The independent check comes from Artificial Analysis, whose Intelligence Index (17 November 2026, maximum effort) puts Astra at 62 — up five points on GPT-6 Astra’s 57, past Google’s Gemini 4 Carbon at 60 and Claude Opus 5.5 at 59, but still behind Claude Fable 5.2 at 65. Sol scores 57, level with the previous GPT-6 Astra, at roughly 15% of its cost per task.
| Model | Artificial Analysis Intelligence Index (17 Nov 2026) | Cost per index task |
|---|---|---|
| Claude Fable 5.2 | 65 | $27.00 |
| GPT-6.2 Astra | 62 | $4.00 |
| Gemini 4 Carbon | 60 | — |
| Claude Opus 5.5 | 59 | — |
| GPT-6.2 Sol | 57 | $1.01 |
| GPT-6 Astra (previous) | 57 | $6.24 |
Here is where the two stories diverge. OpenAI says Astra delivers new highs at less than a third of GPT-6 Astra’s cost per task on its own evaluations. Artificial Analysis measures the cost per task falling 36%, from $6.24 to $4.00, and Vals AI measures a 50% fall, from $21.28 to $10.68. Neither outside tester reaches OpenAI’s figure — but the figures come from different test sets, so none contradicts another outright. A team budgeting from OpenAI’s number would expect a bigger cut than either outside tester found.
There is also a quieter detail inside the independent numbers: on the Artificial Analysis index, Astra generated more tokens per task than its predecessor (65.8 million against 54.3 million), so its lower cost per task comes from the cheaper rates, not from doing less work. Cheaper per token is not the same as cheaper per task — and on a long agentic run, the token count is what you actually pay for.
The safety ratings that matter more than the scores
GPT-6.2 Astra is the first model OpenAI has released externally that it rates High for AI self-improvement under its Preparedness Framework, and both new models are rated Critical for cybersecurity. OpenAI says both models ship with “tighter safeguards for higher-risk cybersecurity assistance”. Do not read the ratings as a safety verdict — they are OpenAI’s own classifications of capability, not a guarantee of safe behaviour.
The context matters: GPT-6 was already rated Critical for cyber capability before this launch, so the 6.2 rating is a continuation, not a new line being crossed. What is new is the self-improvement rating on Astra, which is the part researchers are watching most closely. Early hands-on reaction is mixed — in one small Codex test published on launch day, the tester found no task where Astra won on correctness over Sol — but that is a first impression from a single test, not a verdict.
Which model should you pick?
For most everyday work, Sol is the right default. For the hardest jobs, Astra earns its price — but check the bill, not just the benchmark. OpenAI’s own guidance is the same: Sol for everyday coding and writing, Astra for demanding work such as tracing a bug across services or producing a document that follows a strict template.
| Your job | Pick | Why |
|---|---|---|
| Everyday coding, writing, multi-step tasks | Sol | Within 0.5 points of Astra on FrontierCode 2.0 (OpenAI’s figure) at a fifth of the cost |
| Hardest professional work — finance, law, science | Astra | Highest OpenAI scores on professional-work evaluations; 62 on the independent index |
| Cost-sensitive API use at scale | Sol | $1.01 per Artificial Analysis index task vs $4.00 for Astra |
| Long agent runs in Codex | Either, with Ultrafast | Ultrafast is 50% cheaper for three months; watch the token count, not just the rate |
If you are choosing between Astra and Claude Fable 5.2, the independent index still puts Fable ahead on score (65 vs 62) while Astra costs far less per task ($4.00 vs $27.00 on the same index). That is the real trade-off in this launch: OpenAI is not claiming the top score, it is claiming nearly the top score at a fraction of the price. Whether that trade is worth it depends on whether your work punishes the last three points of capability — for most everyday tasks it does not, which is why Sol, not Astra, is the model OpenAI’s own staff reach for first.
What to watch next
Three things will decide whether GPT-6.2 is a real step or a price cut with a new name. First, whether Sol reaches regular ChatGPT Chat — OpenAI has said nothing about that, and the two models are Work, Codex and API only for now. Second, what happens when the three-month Ultrafast discount ends and Ultrafast pricing steps up. Third, whether independent coding benchmarks show Astra pulling ahead of Sol in real agent work, or whether the two stay as close as the first hands-on tests suggest.
Our take: the honest headline of this launch is not the benchmark table, it is the price sheet. Astra is a better model than GPT-6 Astra and a cheaper one, but the independent cost-per-task saving (36–50%) is roughly half of what OpenAI’s own “under a third” claim implies, and Astra generates more tokens per task than its predecessor. Sol is the quieter story and the better deal for most people: the same score as the old flagship at about 15% of its cost per task. Pick Sol by default, reach for Astra when the job is genuinely hard, and measure your own token bill before trusting anyone’s cost claim — including ours.
Frequently Asked Questions
Are GPT-6.2 Astra and Sol in regular ChatGPT?
No. Both models launched on 17 November 2026 in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, and in the API. An OpenAI staff reply on the developer forum confirms they are not added to regular ChatGPT Chat.
How much cheaper is GPT-6.2 Astra than GPT-6 Astra?
Astra’s standard API rates are $8 input and $40 output per million tokens, 20% below GPT-6 Astra’s $10 and $50, with cached input down 60% to $0.40. Independent testers measured the cost per task falling 36% (Artificial Analysis) and 50% (Vals AI) — less than the “under a third” saving OpenAI claims on its own evaluations.
Which is better, GPT-6.2 Astra or Sol?
Astra scores higher: 62 vs 57 on the Artificial Analysis Intelligence Index. But Sol is within 0.5 points of Astra on OpenAI’s FrontierCode 2.0 test at about a fifth of the cost, and costs $1.01 per index task against Astra’s $4.00. Use Sol for everyday work and Astra for the hardest professional jobs.
Is GPT-6.2 Astra safer than GPT-6?
OpenAI rates Astra as the first externally released model at High for AI self-improvement, and both new models Critical for cybersecurity, with tighter safeguards for higher-risk cybersecurity assistance. These are OpenAI’s own capability classifications, not an independent safety verdict.
When does the Ultrafast discount end?
OpenAI cut Ultrafast pricing by 50% for three months from the 17 November 2026 launch. Pro subscribers at $200 a month get Ultrafast for the first time, starting with Sol; Pro subscribers at $500 a month get it on both models.
Sources and methodology
Claims in this article rest on OpenAI’s launch announcement “Introducing GPT-6.2 Astra and Sol” (17 November 2026, via the OpenAI community post), Artificial Analysis Intelligence Index results dated 17 November 2026, TechTimes’ independent cost comparison (18 November 2026), ArtificialWatch’s GPT-6.2 model page, Metarix’s pricing breakdown, and a hands-on Sol-versus-Astra test published on Medium on 18 November 2026. All benchmark gaps OpenAI quotes are labelled as OpenAI’s own; independent figures are labelled with their source. This article was desk-researched on 18 November 2026 and was not hands-on tested — neither model is available in regular ChatGPT, and OpenAIMaster has no API access to either model. Internal links: our Claude Fable 5.2 explainer, our Gemini 4 Carbon explainer, and our GPT-6.1 Terra launch piece.
- OpenAI launch announcement — https://community.openai.com/t/gpt-6-2-astra-and-sol-are-now-available/1417531/1?utm_source=facebook&utm_medium=paid_social&utm_campaign=120250831925710182&utm_term=120250988291260182&utm_content=120250999307890182&utm_id=120250831925710182
- TechTimes, “GPT-6.2 Astra Cuts Task Costs 36% to 50% in Independent Tests” (18 Nov 2026) — https://www.techtimes.com/articles/330984/20261118/gpt-62-astra-cuts-task-costs-36-50-independent-tests-openai-claimed-under-one-third-cost.htm
- ArtificialWatch, GPT-6.2 Astra model page — https://artificialwatch.com/model/gpt-6-2
- Metarix, “GPT-6.2 Astra Pricing: Astra and Sol API Costs” — https://metarix.io/gpt-6-2-astra-pricing
- Medium hands-on test, “GPT-6.2 Sol vs Astra: Which One Should You Actually Use?” — https://medium.com/@simsketch/gpt-6-2-sol-vs-astra-which-one-should-you-actually-use-0d97e92c63a1
Arva Rangwala covers AI news, AI tools, guides and prompts for OpenAIMaster — what is new, what is worth using, and how to put AI to work.


