OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, a week after Anthropic's Claude Fable 5.2. Astra is the new top model for the hardest projects; Sol is the everyday model most people will actually use. Both are available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers in the API as gpt-6.2-astra and gpt-6.2-sol.
The headline for your wallet: Astra got 20% cheaper than GPT-6 Astra, at $8 input and $40 output per million tokens, and cached input fell 60% to $0.40. Sol kept its price of $2 and $10. OpenAI's claim is that both now finish tasks with fewer tokens, so the real cost per job falls further than the price cut suggests.
Everything below is from OpenAI's announcement and pricing page unless noted. Benchmark numbers are OpenAI's own and have not yet been independently reproduced, with one exception noted.
GPT-6.2 at a glance
| GPT-6.2 Astra | GPT-6.2 Sol | |
|---|---|---|
| Released | November 17, 2026 | November 17, 2026 |
| What it is for | the hardest professional projects, start to finish | everyday work, from coding to running agents at scale |
| API name | gpt-6.2-astra | gpt-6.2-sol |
| Input, per million tokens | $8.00 | $2.00 |
| Cached input | $0.40 | $0.10 |
| Output | $40.00 | $10.00 |
| Change from the previous version | 20% cheaper than GPT-6 Astra; cached input 60% cheaper | same price as GPT-6.1 Sol |
| In ChatGPT | Plus, Pro, Enterprise, Edu (ChatGPT Work and Codex) | Plus, Pro, Enterprise, Edu (ChatGPT Work and Codex) |
| Ultrafast mode | Pro 500 | Pro 500, and for the first time Pro 200 |
There is no GPT-6.2 Luna. OpenAI's light model is still GPT-6 Luna; despite people searching for "GPT 6.2 Luna", only Astra and Sol were announced.
What's new in GPT-6.2 Astra
Office work that reads less like AI. OpenAI's main pitch is professional documents. It says Astra's slides, spreadsheets and documents follow your templates and the conventions of your field more closely, down to the layout, colours and font sizes of a slide template given as a file or even a screenshot, with more direct writing and less filler.
Results on professional benchmarks. On Agents' Last Exam, which tests long, multistep professional work, OpenAI says Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost, and 5.5 points above Claude Fable 5.2 at about 5% of the cost. On Harvey's legal benchmark it says Astra more than doubles Fable 5.2's pass rate at about a sixth of the cost. On its internal GDPval v2, built from real consulting, finance and design projects, Astra improves 6.2 points on GPT-6 Astra while using 22% fewer output tokens.
Science. OpenAI calls it the best model in the world for scientists, with a 9-point gain over GPT-6 Astra on Terminal-Bench-Science at under a third of the cost.
Computer use and 3D. More precise computer use in CAD and simulation tools, and better web and 3D design. On Artificial Analysis's blinded AA-3D comparison, OpenAI reports Astra first at 1,318 Elo, ahead of Fable 5.2 at 1,271 and GPT-6.2 Sol at 1,259.
Independent view. Artificial Analysis, which runs its own index, says Astra moves into second place on its Intelligence Index, behind Claude Fable 5.2.
What's new in GPT-6.2 Sol
Sol is the model OpenAI wants you to use by default. It says Sol is smarter and more efficient than GPT-6.1 Sol, so you complete more tasks within the same paid plan.
- On Cognition's FrontierCode 2.0, which checks whether code changes would be accepted by real maintainers, Sol comes within 0.5 points of Astra at about a fifth of the cost (both at maximum effort).
- On OpenAI's LifeVal, which tests everyday tasks such as shortlisting kindergartens after a move or planning meals for a parent with a new diagnosis, Sol outscores GPT-6 Astra at about a seventh of the cost.
- On GDPval v2, Sol improves 5.3 points on GPT-6.1 Sol.
How it compares on price
| Model | Input / output per million tokens | Notes |
| GPT-6.2 Astra | $8 / $40 | new, 20% below GPT-6 Astra |
| GPT-6 Astra | $10 / $50 | previous OpenAI flagship |
| Claude Fable 5.2 | $8 / $40 | Anthropic's top model since November 9 |
| Gemini 4 Carbon | $2 / $10 until November 30, then $4 / $20 | Google's top model |
| GPT-6.2 Sol | $2 / $10 | new, same price as GPT-6.1 Sol |
| Gemini 4 Hydrogen | $1.50 / $7.50 | Google's cheaper Gemini 4 |
The interesting line is the first and third: OpenAI's and Anthropic's flagships now cost exactly the same per token. The battle has moved from token prices to how many tokens each model needs to finish a job, which is why OpenAI talks so much about "cost per task". Full current prices for 90 models are in our AI model catalogue.
What it means for you
If you pay for ChatGPT Plus ($20 a month), you now get OpenAI's newest frontier model in ChatGPT Work and Codex. For most everyday tasks, Sol is the one to use; switch to Astra for long documents, decks, spreadsheets and hard analysis where quality matters more than speed. If you are on Go or Free, OpenAI's announcement does not list your plan; what each plan includes is in ChatGPT Go vs Plus.
If you build on the API, the arithmetic changed in your favour. A job that cost $10 of input and $50 of output on GPT-6 Astra now costs $8 and $40 at the same token count, and OpenAI says fewer tokens are needed. Cached input at $0.40 instead of $1 rewards keeping your prompt prefixes stable, which OpenAI explicitly recommends. Try Sol first; move to Astra only for the tasks where Sol falls short.
If you are choosing between OpenAI, Anthropic and Google, prices at the top are now level, so test on your own work. Benchmarks from each vendor favour their own model; independent rankings put Fable 5.2 and GPT-6.2 Astra close together. How the assistants differ in daily use is in ChatGPT vs Claude vs Gemini.
Which one to use
| Your task | Pick | Why |
| Emails, summaries, research questions, everyday coding | Sol | close to Astra on many tasks at a fraction of the cost |
| Slides, spreadsheets, financial models, legal or consulting deliverables | Astra | this is where OpenAI says the gains are largest |
| Long agent workflows in Codex | start with Sol, escalate to Astra | Sol is near Astra on FrontierCode at about a fifth of the cost |
| Science, CAD, 3D and simulation | Astra | strongest claims are here |
| High-volume API work | Sol | $2 / $10 versus $8 / $40 |
Safety notes worth knowing
OpenAI says both models meet its "Critical" threshold for cybersecurity capability, so they launch with tighter safeguards that may also block some legitimate security research. Astra is the first model OpenAI has released externally that meets its "High" threshold for accelerating AI research.
One finding is directly useful for everyday users. OpenAI reports that in two of its tests both models were more likely to leave out obstacles that blocked their work, such as a missing file or an unavailable search tool. In practice: when a model says a task is done, check that it really used the files and tools you expected. The habit of checking results is covered in how to write a good prompt.
The honest take
What strikes me about this release is less the benchmarks than the pricing. OpenAI cut its flagship to exactly Anthropic's price, kept its everyday model where it was, and made "cost per task" the number to beat. For ordinary users that is good news: the best models are getting cheaper to run, and the gap between the top model and the everyday one keeps shrinking.
For most people the practical choice is simple: use Sol by default, reach for Astra when the work is long, structured and important. And whatever model you use, check what it actually did.
Frequently asked questions
When was GPT-6.2 released?
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, in ChatGPT Work, Codex and the API.
How much does GPT-6.2 cost?
In the API, GPT-6.2 Astra costs $8 per million input tokens, $0.40 for cached input and $40 for output. GPT-6.2 Sol costs $2, $0.10 and $10. In ChatGPT, both are included in Plus, Pro, Enterprise and Edu plans.
What is the difference between GPT-6.2 Astra and Sol?
Astra is OpenAI's frontier model for the hardest professional projects. Sol is the faster, cheaper everyday model; OpenAI says it comes within half a point of Astra on a coding benchmark at about a fifth of the cost.
Is GPT-6.2 better than Claude Fable 5.2?
OpenAI reports Astra ahead of Fable 5.2 on several of its chosen benchmarks at a lower cost per task. Artificial Analysis's independent index places Astra second, behind Fable 5.2. Both cost $8 / $40 per million tokens.
Is GPT-6.2 available on ChatGPT Plus?
Yes. Both models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Ultrafast mode is available on Pro 500, and on GPT-6.2 Sol for Pro 200.
Is there a GPT-6.2 Luna?
No. Only GPT-6.2 Astra and GPT-6.2 Sol were announced. OpenAI's lightest current model remains GPT-6 Luna.
Sources
- OpenAI: Introducing GPT-6.2 Astra and Sol (checked 17 November 2026)
- OpenAI: API pricing (checked 17 November 2026)
- Artificial Analysis on GPT-6.2 (X)
- 9to5Google: OpenAI releases GPT-6.2 Astra and Sol



