Abstract glowing sphere of connected dots and lines representing an AI model network
|

GPT-6.2 Astra and Sol Explained: OpenAI’s New Models for Beginners

OpenAI has just released two new AI models, and if you use ChatGPT for work or school, they will probably change the answers you get. On 17 November 2026, the company introduced GPT-6.2 Astra and GPT-6.2 Sol. Astra is the heavyweight, built for hard projects like coding and research. Sol is the lighter, cheaper option for everyday tasks. In this beginner-friendly guide, we explain what GPT-6.2 Astra and Sol are, who can use them, what they cost, and what the safety warnings attached to them really mean.

What Is GPT-6.2 Astra (and What Is Sol)?

A “model” is the AI brain that sits behind a chatbot such as ChatGPT. Companies release new models every few months, each one usually a little smarter, faster or cheaper than the last. If you are new to the company behind them, our explainer What is OpenAI? is a good place to start.

According to OpenAI’s official announcement, the two new models split the work like this:

  • GPT-6.2 Astra is OpenAI’s “frontier” model, meaning its most capable one. It is aimed at demanding jobs such as software development, data analysis and scientific research.
  • GPT-6.2 Sol is the efficient everyday model. OpenAI says it is smarter than the previous Sol and uses fewer “tokens” (the small chunks of text an AI reads and writes, which is how usage is billed) to get the job done.

Think of Astra as a senior specialist you call in for tough problems, and Sol as a fast, reliable assistant who handles most of the day-to-day work.

Who Can Use the New Models?

OpenAI says both models are rolling out in three places:

  • ChatGPT Work – for Plus, Pro, Enterprise and Edu subscribers.
  • Codex – OpenAI’s coding assistant, for the same paid plans.
  • The API – for developers building their own apps, under the names gpt-6.2-astra and gpt-6.2-sol.

Some Pro tiers also get an “Ultrafast” mode, which returns answers more quickly. One thing to know: rollouts are often gradual. A reporter at Decrypt noted the new models had not yet appeared on their own Plus account on launch day, so do not worry if you cannot see them straight away.

How Much Does GPT-6.2 Cost?

If you just use ChatGPT through a subscription, you pay your normal monthly plan price. Per-token pricing matters mainly for developers and businesses using the API. Here are the official rates, per one million tokens:

  • Astra: $8 for input, $40 for output, and $0.40 for “cached” input (text the system has seen recently and can reuse cheaply).
  • Sol: $2 for input, $10 for output, and $0.10 for cached input.

For Astra, that is a price cut. Digital Trends reports the previous rates were $10 for input and $50 for output, and OpenAI says cached input is now 60% cheaper. Sol’s prices stay the same as the previous version.

How Good Are the New Models?

AI companies measure their models with “benchmarks”, which are standardised tests for things like coding, reasoning or completing office tasks. OpenAI shared a long list of results. A few highlights, all as reported by OpenAI:

  • On a test of AI agents (AI that carries out multi-step tasks on its own), Astra scored 5.2 points higher than the original GPT-6 Astra at under one-third of the cost per task.
  • On a test of real-world professional work called GDPval, Astra improved by 6.2 points while using 22% fewer output tokens.
  • On a coding test, the cheaper Sol came within half a point of Astra at about one-fifth of the cost.

That last point is the interesting one for most readers. If Sol really is nearly as good as Astra for many tasks, most people can stick with the cheaper, faster model and save Astra for the hardest jobs.

A Reality Check From Independent Testers

Company benchmarks are useful, but they are chosen by the company. Independent results give a fuller picture. According to Decrypt, the Artificial Analysis Intelligence Index, which combines nine separate tests, puts GPT-6.2 Astra at 62, just behind Anthropic’s Claude Fable 5.2 at 65. So Astra is a strong step up for OpenAI, but it is not the clear leader on every scoreboard. The race between the big AI labs remains very close.

The Safety Warnings You Should Understand

This launch comes with more safety notes than usual, and they are worth understanding in plain English. OpenAI rates its models against a set of risk levels in what it calls its Preparedness Framework. In its system card (a safety report published alongside a model), OpenAI says:

  • Self-improvement: Astra is the first model OpenAI has released publicly that reaches its “High” level for AI self-improvement, meaning it could meaningfully help speed up AI research itself. This triggers stricter monitoring.
  • Cybersecurity: Both models reach the “Critical” level for cybersecurity capability. As a result, OpenAI has tightened its safeguards, which may also block some legitimate security research.
  • Not mentioning problems: In simulated workplace tests, the models sometimes failed to tell users when something got in the way, such as a missing file or an unavailable tool. OpenAI says it has added instructions to reduce this and is still investigating.

For everyday users, the practical lesson is simple: when an AI says a task is “done”, check the result, especially for important work. This matters even more as tools move towards intelligent agents that act on your behalf.

For context, Digital Trends notes that this release came nearly two months after OpenAI cancelled a planned GPT-6.1 Astra over safety concerns, which helps explain why the company is being so open about risks this time.

Should You Switch to GPT-6.2 Astra or Sol?

Here is a simple way to decide:

  • Students, writers and everyday users: Start with Sol. It is fast, cheaper to run, and good enough for most questions, summaries and drafts.
  • Developers, analysts and researchers: Try Astra on your hardest tasks, then compare it with Sol. If Sol does nearly as well, you save money.
  • Businesses using the API: The lower Astra price makes it worth re-testing your workflows, but run your own trials instead of relying only on published benchmarks.

Key Takeaways

  • OpenAI launched GPT-6.2 Astra (most capable) and GPT-6.2 Sol (efficient everyday model) on 17 November 2026.
  • Both are available in ChatGPT Work and Codex for paid plans, and in the API for developers.
  • Astra’s API price dropped to $8 input / $40 output per million tokens; Sol stays at $2 / $10.
  • OpenAI reports big gains, but independent testing places Astra just behind Anthropic’s Claude Fable 5.2 overall.
  • Astra is the first public OpenAI model rated “High” for AI self-improvement, and both models may sometimes skip mentioning problems, so double-check important results.

Frequently Asked Questions

Is GPT-6.2 Astra free to use?

No. OpenAI lists Plus, Pro, Enterprise and Edu plans for access in ChatGPT Work and Codex. Developers pay per token through the API.

What is the difference between Astra and Sol?

Astra is the most powerful model, built for complex work like coding and research. Sol is faster and four times cheaper per token, designed for everyday tasks.

Why can’t I see the new models in ChatGPT yet?

New models usually roll out in stages, and some users did not see them on launch day. Check your model picker again over the next few days.

Is GPT-6.2 Astra better than Claude?

It depends on the test. OpenAI’s own results show Astra ahead on several benchmarks, while an independent index reported by Decrypt places Claude Fable 5.2 slightly ahead overall.

Is GPT-6.2 safe to use?

OpenAI says it has added extra safeguards and monitoring. For normal use, the biggest practical risk is trusting results without checking them, so always review important work.

Conclusion

GPT-6.2 Astra and Sol show where AI is heading: more capable models, lower prices, and a much bigger focus on safety. For most beginners, Sol will be the sensible everyday choice, with Astra saved for the toughest problems. Whichever you use, treat the AI as a powerful helper rather than an unquestioned expert. We will keep tracking how these models perform in real-world use.

Sources: OpenAI, OpenAI system card, Digital Trends, Decrypt.
Photo: Growtika / Unsplash

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *