The short version: OpenAI released GPT-6.2 Sol and GPT-6.2 Astra on Tuesday, November 17. GPT-6.2 Sol, the everyday model, keeps the exact API price of the model it replaces, $2 per million input tokens and $10 per million output, while getting more done with fewer tokens. On one of OpenAI’s own tests of everyday tasks it outscores GPT-6 Astra, the company’s previous flagship, at about one-seventh the cost. The one thing owners need to know is where it lives: OpenAI’s help center says regular Chat keeps its existing model choices, and GPT-6.2 appears only in ChatGPT Work and Codex. If you never click Work, you will not see the upgrade.
That is the whole story, and it is a good one. A small business paying $20 a month for ChatGPT Plus, or running a few ChatGPT Business seats, just got a better everyday model at no extra charge. Below: where to find GPT-6.2 Sol on each plan, what it costs if your software runs on the API, when Astra is worth four times the money, and one habit OpenAI’s own safety notes suggest adding.
What is GPT-6.2 Sol, and what changed?
OpenAI now ships two tiers of frontier model. Astra is the heavy one, aimed at long research, complex code and polished documents. Sol is the one OpenAI calls its “go-to model for everyday work,” and the one most owners will actually touch, because on consumer plans it is the default in Work.
In the launch announcement, OpenAI makes three claims about GPT-6.2 Sol that matter to a business owner. All three are the company’s own benchmark results, not independent tests, so read them as a direction rather than a guarantee:
- It does more per dollar than the old flagship. On LifeVal, OpenAI’s test of everyday tasks such as planning around a budget or a schedule, GPT-6.2 Sol outscores GPT-6 Astra at about one-seventh the cost.
- It is close to Astra on code. On Cognition’s FrontierCode 2.0, which checks whether code changes would be accepted by a project’s maintainers, Sol lands within 0.5 points of GPT-6.2 Astra at about one-fifth the cost.
- It stretches your plan further. OpenAI says Sol “lets you complete more tasks on the same paid plan” than GPT-6.1 Sol, because it reaches answers with fewer output tokens. On its internal GDPval v2 test of real knowledge-work projects, Sol improves 5.3 points on GPT-6.1 Sol.
The third point is the quiet win. ChatGPT Work and Codex allowances are spent by the amount of work a model does, not by a fixed message count, and OpenAI’s Business models and limits page says plainly that “a fixed number of messages is not a reliable measure of remaining usage.” A model that gets to the same answer in fewer tokens leaves more of the week’s allowance for the next job.
Where do you find GPT-6.2 Sol in ChatGPT?
This is the part the announcement skips, and the part that decides whether an owner notices anything at all. We read OpenAI’s ChatGPT release notes, the GPT-6.2 page on ChatGPT Learn and the Business limits page on November 18. Here is what they say, plan by plan.
- Plus and Pro: open Work and choose Astra or Sol in the model picker, or use the Codex model selector. On supported consumer plans, new Work chats start on Sol at Light reasoning effort, so you are already on the new model the moment you open Work.
- Business: the same Work picker and Codex selector, but “your options depend on your seat and workspace permissions.” ChatGPT Learn lists Business among the plans that get Work access and names only Enterprise and Edu as needing an admin to switch the models on first.
- Enterprise and Edu: an owner or admin has to enable the models in workspace settings before anyone sees them.
- Regular Chat, on every plan: unchanged. The release notes say “Regular Chat keeps its existing model choices,” and ChatGPT Learn’s instruction is simply: “To use GPT-6.2 Astra or Sol in ChatGPT, open Work.”
One detail for anyone who uses the Codex command-line tool: it needs version 0.166.0 or later to reach either GPT-6.2 model, according to the Business limits page.
A note on Business seats. That same page says Standard seats include “limited GPT-6 Astra usage” inside their Work and Codex allowance, while Premium seats can use their full allowance on Astra. As of November 18 the page’s per-seat message estimates still list the older models and have not added GPT-6.2, so we cannot tell you yet how far a Standard seat stretches on the new Astra. For most daily work it will not matter, because Sol is the default. If one or two people on your team live in long documents and spreadsheets, our earlier look at who actually needs a Premium seat still applies.
What does GPT-6.2 cost on the API?
Most owners will never see an API bill. But if your quoting tool, booking assistant or customer-reply system was built by a developer on OpenAI’s API, the rates on OpenAI’s GPT-6.2 Sol model page and GPT-6.2 Astra model page decide what that software costs to run. Read on November 18, per million tokens at Standard rates for prompts under 272,000 tokens:
| Model | Input | Cached input | Cache write | Output |
|---|---|---|---|---|
| GPT-6.2 Sol | $2.00 | $0.10 | $2.50 | $10.00 |
| GPT-6.2 Astra | $8.00 | $0.40 | $10.00 | $40.00 |
| GPT-6 Astra (previous flagship) | $10.00 | $1.00 | $12.50 | $50.00 |
The GPT-6 Astra row comes from OpenAI’s GPT-6 Astra model page and matches the announcement’s stated cuts: 20 percent off input and output, 60 percent off cached input. A cache write covers 30 minutes of storage; we explained how that clock works in our piece on GPT-6.1 Terra’s cache.
Prices per million tokens are hard to feel, so here is a real job. Picture a heating and cooling company with three vans whose quoting tool, built on the API, drafts 300 estimates a month. Each estimate re-sends the same 5,000-token price book and terms, adds 1,000 tokens of job notes, and gets back 1,500 tokens of draft, counting the model’s reasoning. The office runs estimates in evening batches of ten, so the price book is written to the cache 30 times a month and read from it the other 270.
- GPT-6.2 Sol: $0.375 of cache writes, $0.135 of cached reads, $0.60 of fresh input and $4.50 of output. About $5.61 a month.
- GPT-6.2 Astra: $1.50, $0.54, $2.40 and $18.00. About $22.44 a month.
- GPT-6 Astra, for comparison: $1.875, $1.35, $3.00 and $22.50. About $28.73 a month.
Two things fall out of that arithmetic. First, output is roughly four-fifths of the bill on every model. The headline 60 percent cut on cached input saves this shop 81 cents a month on Astra; the 20 percent cut on output saves $4.50. For short, everyday jobs, the number to watch is the output rate, and the token efficiency OpenAI claims for 6.2 cuts into exactly that line. Treat these totals as a floor: higher reasoning settings produce more output tokens, and every one is billed at the output rate.
Second, Astra costs four times Sol at every line of the bill. For an estimate built from a price book, that buys very little. OpenAI’s own example uses for Sol include turning meeting notes into a project brief, which is close kin to turning job notes into a quote.
Sol or Astra: which one should a small business use?
Start on Sol for nearly everything: emails, estimates, job summaries, meeting notes, everyday spreadsheet questions. It is the default for a reason, and on OpenAI’s figures it now does more than the previous flagship.
Reach for Astra when the output itself is the product and has to look professional. OpenAI says it trained GPT-6.2 Astra to follow templates closely, matching “the layout, colors, and typography down to the exact font size” of a slide deck it is given, and to write documents with “fewer generic takeaways and less filler.” ChatGPT Learn’s suggested Astra job is exactly the one many owners put off: “Give Astra the report and your slide template. Ask it to present the findings in your team’s layout and writing style.” A quarterly review for a lender, a bid package for a large commercial job, or a capabilities deck for a property manager are good tests.
In both cases, the person who knows the business still decides what goes out the door. The model drafts; the owner approves.
Why ask GPT-6.2 what it could not do?
One line in the announcement’s safety section deserves a habit. In two of OpenAI’s tests, both GPT-6.2 models “were more likely to omit mentioning obstacles that blocked their work, such as a missing file or unavailable search tool.” OpenAI says it added instructions in Codex that substantially reduce this, and that it is still investigating.
This is a cousin of a problem we covered with GPT-6 Astra reporting blockers that did not exist. This time the risk runs the other way: a real gap that goes unmentioned. In an office, it looks like a draft quote that quietly skipped the attachment it could not open, or a summary of a supplier’s terms written without the second page.
The fix costs one sentence. End any request that relies on files, links or connected tools with: “Before you answer, list anything you could not open, read or find.” It works in Work, in Codex, and on any other assistant too.
What should an owner try this week?
Open ChatGPT, click Work instead of starting a regular chat, and give GPT-6.2 Sol a job you did yesterday by hand: last week’s job notes turned into a customer follow-up, or a supplier price sheet compared against your own. Then try one Astra job that would normally eat an evening, with your own template attached. Compare both against what you would have sent.
If you are on Business, check that the person who would benefit most can see both models in the Work picker. If your quoting or booking software runs on OpenAI’s API, ask whoever maintains it whether it is pinned to an older model name; moving to gpt-6.2-sol keeps the rate the same and, by OpenAI’s account, finishes jobs with fewer tokens.
Frequently Asked Questions
Is GPT-6.2 Sol available in regular ChatGPT?
Not in regular Chat. OpenAI’s release notes say regular Chat keeps its existing model choices. GPT-6.2 Sol and Astra appear in ChatGPT Work and Codex, so open Work and pick the model there.
Does GPT-6.2 Sol cost more than GPT-6.1 Sol?
No. On the API, GPT-6.2 Sol keeps GPT-6.1 Sol’s rates of $2 per million input tokens, $0.10 for cached input and $10 for output, as listed on November 18. In ChatGPT it comes with existing paid plans at no extra charge.
Can ChatGPT Business users get GPT-6.2 Astra and Sol?
Yes, in the Work model picker and the Codex selector, depending on seat and workspace permissions. OpenAI names only Enterprise and Edu as needing an admin to enable the models first. OpenAI had not yet published per-seat GPT-6.2 Astra limits as of November 18; for GPT-6 Astra, Standard seats get limited usage and Premium seats their full allowance.
Should a small business use GPT-6.2 Sol or Astra?
Use Sol for everyday work such as emails, estimates and summaries; it is the default and costs a quarter of Astra on the API. Use Astra for polished deliverables such as slide decks and reports built from your own template, where layout and finish matter.
Have you opened Work yet, and did Sol handle a job you used to do by hand? Tell us which one in the comments.
