Have you ever asked an AI for a slide deck or a budget spreadsheet and gotten something that looked generic and needed a lot of cleanup? That's the problem OpenAI says these new models are designed to solve.
Two models, two jobs
OpenAI describes two versions in its family:
- GPT-6.2 Astra is the company's new top-tier model, built for demanding work. OpenAI says its slides, spreadsheets, and documents read less like AI-written text and follow the templates and conventions of your field more closely. It also says Astra writes cleaner code and builds better-looking websites and 3D designs.
- GPT-6.2 Sol is the everyday model. Compared with GPT-6.1 Sol, OpenAI says it is smarter and more efficient, and it lets you complete more tasks on the same paid plan. It is also built to handle multistep workflows, which are tasks where the AI carries out several connected steps in a row.
Think of Astra as the specialist you bring in for a big, complicated project, and Sol as the reliable colleague you go to for daily work.
Why "fewer tokens" matters
OpenAI emphasizes that both models produce better results with fewer output tokens. A token is a small chunk of text, often a word or part of a word, that an AI reads or writes. Output tokens are the pieces the AI generates while it reasons and writes its answer.
Why should you care? Fewer tokens can mean you hit your usage limits more slowly, and for businesses using the API (the behind-the-scenes connection that lets programs talk to AI), it can affect cost at scale. OpenAI frames this as getting more done for the same effort.
Astra for Law: a setup built for lawyers
OpenAI has also released an update to Astra for Law, a configuration of GPT-6.2 Astra designed for legal work. OpenAI says it pairs the model with an expanded legal search index (a larger library of legal sources to look things up in) and instructions for researching a question, weighing the sources, and writing an answer.
OpenAI reports that on Harvey LAB-AA, a test of legal assignments, Astra for Law earned full credit on 20.8% of tasks, compared with 15.0% for GPT-6.2 Astra. A full-credit answer has to meet every requirement without a hallucination, which is when an AI confidently states something its source documents don't support. OpenAI also reports 0.03 material hallucinations per assignment in its own evaluation. These are the company's own test results, and you can read the official post for full details.
Access works differently from the standard ChatGPT tools. OpenAI says all ChatGPT Enterprise customers can now request access through its Trusted Access Program, including through the API. If your workplace uses an Enterprise plan, your administrator is the best person to ask about it.
Wrap-up
GPT-6.2 Astra targets demanding professional work, GPT-6.2 Sol targets everyday tasks, and Astra for Law brings a legal-focused setup to ChatGPT Enterprise customers who request access. Most of these claims come from OpenAI's own tests, so treat them as the company's view until you see results in your own work. Your next step: pick one document you write often, and this week, compare how well your current AI tool follows its format, then check the official OpenAI announcement for which models your plan includes.
