OpenAI's New GPT-6.1 Sol Costs a Fifth as Much. What It Means

At its DevDay event on 29 September, OpenAI launched GPT-6.1 Sol, which it says nearly matches its top model, GPT-6 Astra, on coding, computer use and professional work, at one-fifth of Astra's token prices. OpenAI's announcement is the primary source, and TechCrunch, The Next Web and MarkTechPost all report the same core facts.

If you use AI at work, or pay someone who does, this is the kind of change that quietly alters what's affordable.

What actually happened

According to the coverage, GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and in Codex, and to developers through the API. TechCrunch notes it is not yet in the main Chat experience.

On price, the outlets report API rates of $2 per million input tokens and $10 per million output tokens, one-fifth of Astra's standard rates. MarkTechPost reports Astra stays on as a premium tier at $10 and $50, which OpenAI recommends for "the hardest research" work. A "token" is roughly a small chunk of a word, so these are the prices developers and software vendors pay, not what a ChatGPT subscriber sees on a bill.

On quality, these are OpenAI's own claims: it says Sol "approaches GPT-6 Astra's performance" on several fronts, and TechCrunch reports OpenAI's figures showing factual error rates falling from 11.4% to 7.7% at low reasoning effort. The Next Web reports OpenAI's own benchmarks showing Sol ahead of Anthropic's Opus 5.5 on one automation test at about a third of the cost.

TechCrunch also reports that OpenAI did not launch GPT-6.1 Astra as expected. It attributes to the Wall Street Journal that the company scrapped it over safety concerns after internal testing, including a tendency to proceed with tasks without asking permission.

What it means for a small business or recruiter

Cheaper models change the economics, not the nature, of AI tasks. Work that was too expensive to run at volume, such as screening large numbers of CVs, drafting first versions of job ads, summarising call notes or checking documents, gets cheaper to try. Software vendors who build on these models may also pass on savings, or not. That's worth asking about at renewal.

It also keeps the pace of change high: the model you evaluated three months ago is probably not the best option now.

What to do

1. Check what you're actually paying for. If you're on a Plus, Business or Team-style plan, look in ChatGPT for the new Work area and try Sol on a task you already do.
2. Test with a real task, not a demo. Take one job you do weekly and compare the output with what you get today.
3. Ask your software vendors whether and when cheaper models will reduce your costs or improve what you get.
4. Keep a human check on anything sent to clients or candidates, particularly for AI that acts on its own.
5. Don't rebuild processes around one model. Keep your workflows portable.

The caveat

Almost everything on performance and price comes from OpenAI's announcement and its own benchmarks, which haven't been independently verified in the coverage we read. Cheaper per token doesn't guarantee a cheaper bill: long documents and heavy "reasoning" settings use many more tokens. Plan names, limits and availability are changing quickly, so check your own account. The Astra safety detail is reported via the Wall Street Journal, and we haven't seen OpenAI's own account of it.

If you'd like help working out where AI fits into your recruitment workflow, get in touch.