GPT-6.2 Astra and Sol: what OpenAI's new models mean at home

OpenAI's GPT-6.2 Astra and Sol are aimed at work, coding and computer use, not home devices. Here is who can use them, what they cost, what to check and what to watch next.

home, interiors, kitchen, kitchen counter
Image: user32212 on Pixabay

OpenAI has released two new models, GPT-6.2 Astra and GPT-6.2 Sol, for ChatGPT Work, Codex and, according to its announcement, the API. They are pitched at professional documents, coding and computer use. Our news story on the launch covers what happened. This piece covers what it means if you run a smart home, solar system or EV.

The short answer: very little changes on your devices today. Nothing in the material we reviewed describes a new smart home feature, device integration or home assistant function. The models run in OpenAI's own tools, not in your hub, speaker or car.

The launch still matters indirectly. If you use ChatGPT to plan projects, write Home Assistant scripts or tidy paperwork, your plan decides whether you get these models. If you pay for AI through the API, the prices change the sums. Much of the performance evidence is still OpenAI's own.

What are GPT-6.2 Astra and Sol?

They are two new AI models from OpenAI. A model is the engine behind a chatbot or coding tool; the app you type into is just the front end. According to 9to5Google's report, OpenAI describes Sol as the model for everyday work: smarter and more efficient, so people can get more done within their current subscription.

Astra is aimed at the highest-quality professional output. That includes documents, slides, spreadsheets, coding, and front-end design for websites, games and 3D projects. OpenAI says the output follows the templates you supply and the conventions of your field.

The announcement also stresses stronger "computer use". That means an AI working a computer the way a person does: reading the screen, clicking and typing to finish a task. OpenAI's announcement says the models bring this, along with better coding and lower cost.

Does this change anything in my smart home today?

Not directly. OpenAI's announcement, as we summarised it in our main story on the launch, does not mention smart home features, device integrations or a home assistant function. The models are offered in ChatGPT Work, Codex and the API.

That means your lights, thermostat, inverter and charger will behave exactly as before. A speaker or hub only gets these models if its maker chooses to connect to them through the API, and no source we reviewed says anyone has done so.

The effect, if any, arrives in the apps and services built on top of the models. Our page on AI at home tracks those as they appear.

Who can use Astra and Sol right now?

9to5Google reports that both models are available now to people on Plus, Pro, Enterprise and Edu plans, in ChatGPT Work and in Codex. If you pay for one of those plans, check the model picker in the app rather than assuming it has changed.

The sources disagree on, or leave open, several access questions. OpenAI's announcement, as we read it, lists the API as a third place for the models, but gives no exact dates for each product. 9to5Google says it does not know whether the models will reach free users or the API, and gives no detail on how long they will be available. Treat those points as unconfirmed until OpenAI or a plan page says otherwise.

The table below sets out what has been reported for each model.

What has been reported about GPT-6.2 Astra and Sol (sources: 9to5Google, OpenAI announcement)
SolAstra
Pitched asEveryday work, more done within your plan
Typical tasks namedDay-to-day work
Input price per million tokens$2
Cached input per million tokens$0.10
Output price per million tokens$10
Plans reportedPlus, Pro, Enterprise, Edu

What do the API prices mean in practice?

Developers and tinkerers who pay per use are billed in tokens, which are small chunks of text, roughly part of a word each. You pay one rate for what you send in (input), a lower rate for repeated material the service has stored (cached input), and a higher rate for what the model writes back (output).

As a rough illustration using the reported prices, a job that sends 100,000 tokens and gets 10,000 back would cost about $0.30 on Sol ($0.20 in, $0.10 out) and about $1.20 on Astra ($0.80 in, $0.40 out). Those are our own sums, not figures from OpenAI. Real costs depend on how much text your task actually uses.

According to 9to5Google, Sol keeps the price of its predecessor, while Astra is significantly cheaper than the earlier GPT-6 Astra. That is the report's wording; we have not seen the old price list. Also note that OpenAI's announcement, as we summarised it, does not carry a full price list, so the numbers come from 9to5Google. Check OpenAI's own pricing page before you build anything around them.

How good are they compared with rivals?

OpenAI's own comparisons, as relayed by 9to5Google, show the models performing close to Anthropic's latest, often at a much lower cost. The report notes these are the company's figures and includes no independent testing.

The context is a crowded fortnight. 9to5Google says the launch came about a week after Anthropic released Claude Fable 5.2 and Mythos 5.2, and after Google's Gemini 4 Hydrogen. Anthropic had said Fable 5.2 beat rival models across most benchmarks in coding, cybersecurity and knowledge work. For the Anthropic side, see our explainer on Claude Fable 5.2 and what cheaper, faster AI means at home.

Other outlets have covered the launch with different emphasis. Decrypt's headline says Astra challenges Anthropic on price but not on every benchmark, which fits the picture of a cheaper model that does not win across the board. Tom's Guide published a hands-on piece in which its writer built a website from a single prompt; that is the outlet's own experience, not ours. TechRadar's headline highlights a browser-based recreation of Paris as an example of the models' power. Simon Willison, an independent developer who writes about AI models, has a write-up as well, and Digital Trends and 9to5Mac also covered the launch. Read several, since a single demo says little about your own tasks.

Is it worth using for Home Assistant scripts and home projects?

It may be, but nothing we have seen proves it. The sensible home uses of a work model are the dull ones: drafting an automation, explaining an error message, summarising a manual, comparing tariffs in a spreadsheet, or sorting household paperwork.

The risk with AI-written automations is the same as ever. A script that looks tidy can still do the wrong thing, such as switching a heater on at the wrong time. Review what it writes, and test changes on one non-critical device first. Keep anything safety-related, like heating limits, door locks and alarms, out of unattended AI control.

Computer use deserves extra caution. Letting an AI click through accounts on your behalf, say your energy supplier's portal or your router settings, gives it real power over real things. We would not hand that over for anything you cannot undo until independent testing exists.

For the wider picture, our AI roundup for week 46 looks at how assistants are moving into cars, bank accounts and documents, and gives a simple check for each.

What should I do now?

Most readers need only a few minutes. Work through these in order.

  1. Check your plan. If you are on Plus, Pro, Enterprise or Edu, look for Astra and Sol in ChatGPT Work or Codex. If you are on a free plan, do not expect them yet.
  2. Do not upgrade just for these models. With benchmarks coming mainly from OpenAI, wait for independent tests before paying more.
  3. If you use the API, read OpenAI's own pricing page and set a spending cap before running anything automated.
  4. If you use an AI assistant tied to your home, watch whether its maker adds the new models. Read the changelog and the privacy terms when it does.
  5. Keep a human in the loop for anything that controls heating, security or money.

If you want to compare a different rival first, our look at Gemini 4 Carbon and what it means at home covers Google's model aimed at security work and long, multi-step tasks.

What should I watch next?

Four things will tell you whether this launch ever reaches your home.

  • Independent tests. Third-party benchmarks and real-world comparisons against Claude Fable 5.2 and Google's latest models, especially for coding and spreadsheets.
  • API access and dates. Confirmation of exactly when and where each model arrives, and whether free users get any access.
  • Third-party adoption. Whether smart home hubs, assistants and automation tools add Astra or Sol. No source we have confirms that yet.
  • Pricing moves. Rivals have been cutting prices, so rates could shift again within weeks.

Until then, the practical reading is simple. This is a story about work and coding tools. It will reach homes only through the apps that choose to use it, and those apps have yet to say so. Our AI section will follow any that do.

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

According to 9to5Google, Sol is the everyday work model, described as smarter and more efficient. Astra is aimed at the highest-quality professional output, such as documents, slides, spreadsheets and front-end design. Astra also costs more per token in the API, as reported.

Is GPT-6.2 free to use?

Not as far as the sources show. 9to5Google reports availability on Plus, Pro, Enterprise and Edu plans, and says it does not know whether free users will get the models. Check the model picker in your own account.

Can I use GPT-6.2 with Alexa, Google Home or Home Assistant?

Nothing we reviewed says so. OpenAI's announcement does not describe any smart home integration. Any support would depend on those platforms or third-party developers adding the models, and none has been confirmed.

Is GPT-6.2 better than Claude Fable 5.2?

That is not settled. OpenAI's own comparisons put its models close to Anthropic's, often at lower cost, but 9to5Google notes there is no independent testing yet. Anthropic made its own claims for Fable 5.2 in coding, cybersecurity and knowledge work.

RelatedClaude Fable 5.2 explained: what cheaper, faster AI means at homeRelatedAI roundup, week 46 of 2026: assistants move into cars, bank accounts and documents

Enjoying this? Get the AI brief

Every Thursday: AI that matters for your home.

✓ Free · seven editions · one click to leave

Topics
Add us as a preferred source on GoogleSee our stories first in Google’s Top Stories.
Get alerts on big newsOnly the big news, at most once a day. No email, no account.