GPT-6.2 Astra and Sol: price, access and what's new

OpenAI's GPT-6.2 launch artwork with the words Astra, GPT-6.2 and Sol in white on a dark navy sky, between two curved trails of sparkling stars and a glowing orange sun

/ OpenAI's GPT-6.2 Astra and Sol are out in ChatGPT Work and Codex. Who gets them, Plus limits, API prices, what's new and what early testers found.

by Hozefa Khety

· 6 min read

OpenAI has two new AI models. GPT-6.2 Astra, its most powerful model, and GPT-6.2 Sol, its cheaper everyday model, launched on Tuesday, November 17, 2026, in ChatGPT Work, Codex and the OpenAI API.

The headline for developers is price: GPT-6.2 Astra costs 20% less per token than GPT-6 Astra, while Sol keeps its old rates. The catch for regular ChatGPT users is that you need a paid plan, and early reviews from some well-known testers are harsh. Here is who can use the models, what they cost, what OpenAI says is better and what independent testers found. Details checked 18:55 IST (13:25 UTC), November 18.

Who can use GPT-6.2 Astra and Sol

OpenAI says both models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI's ChatGPT help pages also list the Business plan. Enterprise and Edu workspaces get the models switched off by default, so an administrator has to turn them on in the workspace model settings.

You will not find them in the regular Chat side of ChatGPT. To use GPT-6.2 Astra or Sol, open Work on the web, desktop or mobile app, or pick them in the Codex app and command-line tool. On consumer plans, new Work chats start on Sol at its lightest reasoning setting, and you can open Advanced to choose Astra or a deeper effort level.

Free and Go plans are not listed. OpenAI's Codex pricing page shows those plans using the smaller GPT-6 Luna model instead.

OpenAI graphic of a white Chat and Work toggle switch with Work selected, on a dark starry background
GPT-6.2 Astra and Sol live in the Work side of ChatGPT and in Codex, not in regular chat. (Image: OpenAI)

How much use you get on Plus and Pro

ChatGPT Work and Codex share one usage allowance. OpenAI's own estimate for Plus and standard Business users is about 6 to 50 messages every five hours with GPT-6.2 Astra, and about 17 to 180 with GPT-6.2 Sol. The range is wide because big, long-running tasks use far more of the allowance than quick ones, and weekly limits can also apply.

That is slightly more than the older models: OpenAI lists 5 to 45 for GPT-6 Astra and 15 to 160 for GPT-6.1 Sol. OpenAI says Pro plans currently have no five-hour limit.

ChatGPT Pro comes in three tiers at $100, $200 and $500 a month. Pro $500 subscribers can use Ultrafast, OpenAI's quickest response mode, with either new model. For the first time, Pro $200 subscribers also get Ultrafast, but only with GPT-6.2 Sol.

GPT-6.2 Astra and Sol API prices

Developers can call the models as gpt-6.2-astra and gpt-6.2-sol in the OpenAI API, with Ultrafast available for both from day one. These are OpenAI's standard rates per million tokens (a token is roughly a short word or part of a word):

GPT-6.2 Astra: $8 for input, $0.40 for cached input and $40 for output. GPT-6 Astra cost $10, $1 and $50, so that is 20% off input and output and 60% off cached input.

GPT-6.2 Sol: $2 for input, $0.10 for cached input and $10 for output, the same as GPT-6.1 Sol.

OpenAI's pricing page also lists Batch processing at half price (Astra $4 and $20, Sol $1 and $5) and a Fast tier at double the standard rate (Astra $16 and $80, Sol $4 and $20). Data residency is available in the US and EU for Standard, Fast and Batch, while Ultrafast is US-only.

For comparison, VentureBeat notes that Anthropic's Claude Fable 5.2, launched eight days earlier, also costs $8 and $40 per million input and output tokens, and Google's Gemini 4 Carbon launched at introductory rates equal to Sol's.

What's new in GPT-6.2 Astra

OpenAI pitches Astra for its users' hardest work. It says the model's slides, spreadsheets and documents read less like AI and follow the templates and conventions people give it more closely, down to matching fonts and layouts from a template file or even a screenshot. It also says Astra writes cleaner code and builds better-looking websites, games and 3D scenes.

OpenAI's headline claims: on Agents' Last Exam, a test of long, multi-step professional tasks, Astra scores 5.2 points higher than GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at about 5% of the cost. OpenAI also calls Astra the best model in the world for scientists, pointing to a 9-point gain on Terminal-Bench-Science, a test of scientific work in a computer terminal.

A big part of the pitch is efficiency. OpenAI says Astra gets better results with fewer output tokens, which means more work fits inside a plan's usage limits and each task costs less through the API.

What's new in GPT-6.2 Sol

Sol is the everyday model, and OpenAI says many of its own staff now use it as their default. Compared with GPT-6.1 Sol, OpenAI says it is smarter, more efficient and lets you complete more tasks on the same paid plan.

OpenAI's standout claim is that on FrontierCode 2.0, a test of whether code changes are good enough to be merged into real projects, Sol comes within 0.5 points of Astra at about one-fifth of the cost. On LifeVal, OpenAI's test of everyday tasks such as planning meals or shortlisting homes, it says Sol beats GPT-6 Astra at about one-seventh of the cost.

What independent testers found

Outside benchmarks mostly back OpenAI on price and performance, with some clear weak spots. On Artificial Analysis's Intelligence Index, which combines nine tests, GPT-6.2 Astra scores 62 and ranks second behind Claude Fable 5.2 (65), while Sol scores 57, matching GPT-6 Astra at about 15% of the cost per task, The Decoder reported.

Coding is where the gains stall. Artificial Analysis found Astra flat on its Coding Agent Index compared with GPT-6 Astra, and Sol three points lower than GPT-6.1 Sol. It also measured high hallucination rates on very hard factual questions at maximum effort (59% for Astra, 55% for Sol), meaning the models often guess instead of admitting they don't know.

Research group Epoch AI rated Astra level with Claude Fable 5.2 on its overall capabilities index, but said it fell behind Fable on real research tasks. Vals AI ranked Astra 7th and Sol 9th on its index, with Sol the cheapest model in its top 10 at $3.18 per task against Astra's $10.68, though both ranked much lower on legal research.

Some early testers were blunt. As reported by VentureBeat, investor Matt Shumer said the models were "essentially unusable" and developer Theo Browne called 6.2 Astra "the worst model OpenAI has ever released". Wharton professor Ethan Mollick had a mixed experience, finding Astra "very uneven & weird" and harder to steer than rivals, while Every CEO Dan Shipper said it often stopped early on end-to-end builds and could be too brief in its writing.

Safety limits you might notice

OpenAI says both models reach its Critical level for cybersecurity ability and its High level for biological and chemical ability, so they launch with tighter safeguards. OpenAI admits these also block some legitimate security research and testing work.

GPT-6.2 Astra is also the first model OpenAI has released publicly that meets its High threshold for AI self-improvement, meaning it can meaningfully speed up AI research. OpenAI says it has added more monitoring and controls because of that.

One quirk is worth knowing if you use the models for agent-style tasks. OpenAI says that in two of its tests, the new models were more likely to leave out obstacles that blocked their work, such as a missing file or a search tool that wasn't working. It has added instructions in Codex to reduce this and says it is still investigating.

Should you switch to GPT-6.2?

If you pay for ChatGPT Plus or Pro and already use Work or Codex, there is little reason not to try GPT-6.2 Sol: it costs nothing extra, OpenAI's estimates give slightly more messages per five hours than GPT-6.1 Sol, and you can still choose a different model under Advanced if you prefer.

For developers, Astra is now cheaper per token than GPT-6 Astra and priced the same as Claude Fable 5.2, so it is worth testing on your own tasks. Given the split early reviews and the flat coding scores from Artificial Analysis, test before moving important workloads. On the free plan, nothing changes for now.

Top image: OpenAI's GPT-6.2 Astra and Sol launch artwork. (Image: OpenAI)

OpenAIGPT-6.2GPT-6.2 AstraGPT-6.2 SolChatGPTCodexAI modelsOpenAI APIClaude Fable 5.22026

See more of The Threshold on Google

Add us as a preferred source to see more of our stories in your Top Stories.

Frequently asked questions

When did GPT-6.2 Astra and Sol launch?

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, November 17, 2026, in ChatGPT Work, Codex and the OpenAI API.

Is GPT-6.2 free in ChatGPT?

No. GPT-6.2 Astra and Sol are for paid plans: Plus, Pro, Business, Enterprise and Edu. OpenAI does not list the Free or Go plans.

Where do I find GPT-6.2 in ChatGPT?

Open the Work side of ChatGPT on the web, desktop or mobile app, or use the Codex app and CLI. The models are not in the regular Chat model list.

How much does GPT-6.2 Astra cost in the API?

$8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens at standard speed, 20% less than GPT-6 Astra for input and output.

How much does GPT-6.2 Sol cost in the API?

$2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens, the same as GPT-6.1 Sol.

How many GPT-6.2 Astra messages do Plus users get?

OpenAI estimates about 6 to 50 local messages every five hours with Astra and 17 to 180 with Sol on Plus, depending on task size. Weekly limits may also apply.

What is the difference between GPT-6.2 Astra and Sol?

Astra is OpenAI's most powerful model for complex research, coding and documents. Sol is cheaper and faster for everyday coding, writing and multi-step tasks, and OpenAI says it comes close to Astra on some coding tests.

Is GPT-6.2 Astra better than Claude Fable 5.2?

It depends on the test. OpenAI says Astra beats Fable 5.2 on Agents' Last Exam at far lower cost, but Artificial Analysis ranks Fable 5.2 slightly higher overall, and the two cost the same per token.

Related