Home » News » OpenAI’s GPT-6.2 Astra and Sol are in ChatGPT Work and Codex now. Here’s who gets them and what they cost

OpenAI’s GPT-6.2 Astra and Sol are in ChatGPT Work and Codex now. Here’s who gets them and what they cost

OpenAI chart of GPT-6.2 Astra and Sol scores against cost on the FrontierCode 2.0 coding test

If you pay for ChatGPT, two new OpenAI models arrived on Nov. 17, called GPT-6.2 Astra and Sol. OpenAI says they are available now in ChatGPT Work and Codex, its coding tool, for Plus, Pro, Enterprise and Edu users. The announcement does not mention the regular ChatGPT chat window, and free accounts are not on the list.

The short version

OpenAI released GPT-6.2 Astra for hard projects and GPT-6.2 Sol for everyday work, for paying ChatGPT users and developers.

  1. Plus, Pro, Enterprise and Edu users get both models in ChatGPT Work and Codex starting Nov. 17, 2026.
  2. OpenAI says Astra sets new highs on its tests at less than a third of the cost per task of GPT-6 Astra.
  3. For developers, Sol costs $2 per million input tokens and $10 per million output tokens, one quarter of Astra’s $8 and $40.
  4. Pro 500 subscribers get the faster Ultrafast mode on both models, and Pro 200 subscribers get it on Sol first.
  5. The downside: OpenAI added tighter limits on higher-risk cybersecurity help, so some security questions may be refused.

What changes for you: on a paid plan, you now have two clear choices inside ChatGPT Work and Codex. Sol is the one OpenAI built for everyday writing, planning and multistep tasks. Astra is for the big jobs, such as a full slideshow, a website or a 3D design. If you are on the free plan, nothing changes today.

Who gets GPT-6.2 Astra and Sol

Here is what OpenAI’s announcement says each group gets, starting Nov. 17:

  • ChatGPT Plus, Enterprise and Edu: both models in ChatGPT Work and Codex.
  • ChatGPT Pro 200: both models, plus Ultrafast with GPT-6.2 Sol. OpenAI says this is the first time Pro 200 gets Ultrafast.
  • ChatGPT Pro 500: both models, with Ultrafast on either one.
  • Developers: gpt-6.2-astra and gpt-6.2-sol in the OpenAI API, with Ultrafast available.
  • Free accounts: not named in the announcement.

Our explainer on GPT-6.1 Terra, the model that clicks and types in your apps covers the last big upgrade for paid plans, and our GPT-6 Luna explainer covers what free accounts got in October.

GPT-6.2 Astra vs Sol: which one to use

OpenAI describes Astra as “a frontier model built for your hardest projects.” It says you will see the difference in “things like slideshows, websites, and 3D designs.” On a test of AI agents called Agents’ Last Exam, OpenAI says Astra scored 5.2 points higher than GPT-6 Astra at less than half the cost per task.

OpenAI chart of Agents’ Last Exam scores against cost for GPT-6.2 Astra, GPT-6.2 Sol, GPT-6 Astra and other AI models
Image: OpenAI

Sol is the everyday model. OpenAI says it “writes clearly and naturally” and handles tasks with several steps. On a coding test called FrontierCode 2.0, OpenAI says Sol nearly matched Astra at about one-fifth of the cost.

A simple way to choose: start with Sol for emails, summaries, planning and questions. Switch to Astra when the job is large and the result has to look finished, like a presentation you will show at work. Keep in mind that these scores come from OpenAI’s own tests.

Insights by TechCity, every Sunday.

Tech explained for people who use it.

What the new models cost developers and small businesses

If your business pays for OpenAI’s API, the price gap matters more than the test scores. These are OpenAI’s prices per million tokens. A token is a small piece of a word.

  • GPT-6.2 Astra: $8 for input and $40 for output.
  • GPT-6.2 Sol: $2 for input and $10 for output.

We did the math. Sol costs one quarter of Astra for the same number of tokens. A small team that generates 10 million output tokens a month would pay about $400 a month with Astra and about $100 with Sol, before input costs. That is why testing Sol first is the cheaper bet for most work.

The trade-off: tighter limits on security help

OpenAI says GPT-6.2 Astra is its “most capable model ever for cyber,” meaning cybersecurity work. Because of that, both models launch with tighter safeguards on higher-risk security help. OpenAI says it is working to give approved security teams access with fewer refusals through a program called Daybreak.

For most people this changes nothing. If you work in IT or security, expect some requests to be turned down that older models answered.

What we don’t know yet

  • Whether free accounts will get either model later.
  • How many messages each paid plan can send with Astra before hitting a limit.
  • Whether older models will leave the ChatGPT model menu, and when.

If you let ChatGPT act on your behalf, our guide to setting spending limits on AI agents is worth a look before you hand a newer model more to do.

Share

WhatsApp X LinkedIn Email

Insights by TechCity, every Sunday.

Tech explained for people who use it.

Olawale Adeyina Avatar