OpenAI’s new model promises near-Astra intelligence for a fifth of the price. For Europe’s small businesses, the price is the headline. The word “near” is the fine print.
FROM THE EDITORS
OpenAI names its models after the sky. Astra is the stars: distant, brilliant and priced like it. Sol is the sun: the light you actually work by. On the same day it launched dots, its always-on agents running on Astra, OpenAI released GPT-6.1 Sol and told the market that most of the starlight is now available by daylight, at a fifth of the price.
The numbers are simple enough to fit on a Post-it. Through the API, GPT-6.1 Sol costs $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. Astra costs $10 input, $1 cached and $50 output. OpenAI’s pitch is that Sol gets close to Astra on complex professional work, coding and computer use, while Astra stays the model for the hardest jobs.
What a fifth looks like on a small firm’s invoice.
Take a ten-person wholesaler in Aalborg or Ghent that wants a model to read a thousand supplier PDFs a month (price lists, spec sheets, order confirmations) and pull out the numbers. Say each one is 20,000 tokens in and 2,000 out. On our arithmetic, that’s about $60 a month on Sol and about $300 on Astra. Neither figure will break a business. But one is a line item and the other is a meeting.
The more interesting number is the ten cents. Cached input, the part of a prompt you send again and again, costs 95% less than fresh input. For a small business, that repeated part is the valuable part: your price list, your terms of sale, your product catalogue, the house style for customer emails. Load it once and every question against it is close to free. That changes the maths on the boring, repetitive work SMEs actually have.
OpenAI’s own scientific benchmark shows the same shape. On Terminal-Bench Science, Sol averaged $5.47 per task, against $23.21 for Anthropic’s Opus 5.5 and $23.80 for Astra. On the GDP.pdf document benchmark, OpenAI puts Sol at roughly a fifth of Astra’s cost per task and roughly a third of Opus 5.5’s.
Now the fine print.
“Near” is doing a lot of work. These are OpenAI’s benchmarks, chosen and run by OpenAI, and a benchmark is not your workload. The factuality result is honest and worth reading closely: at low reasoning effort, Sol cut the share of answers containing a factual error from 11.4% to 7.7%. That is a real 32% improvement. It is also roughly one answer in thirteen that still gets something wrong. If Sol is pulling prices off supplier PDFs, someone still checks the numbers before they reach a quote.
Where you can use it matters too. In ChatGPT, GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise and Edu users, but only in ChatGPT Work and Codex, not yet in ordinary Chat. Developers get it in the API as gpt-6.1-sol, and a faster Ultrafast version, with up to 8x the token speed in Codex, is promised in the coming days. The launch post says nothing about Europe either way. After dots shut out Pro users in the EEA the same afternoon, European readers will want to check their own accounts rather than take a global launch at its word.
And cheaper doesn’t change who’s responsible. Whether you pay $2 or $10 per million tokens, customer data sent to the API is still personal data you control under GDPR. The price of the model has dropped. Your obligations haven’t.
How to work by sunlight:
- Default to Sol, escalate to Astra. Run the everyday work on the cheap model and send only the tasks it fails to the expensive one.
- Put your fixed material in the cache. Price lists, terms and catalogues are what you send every time, and that’s where the 95% discount sits.
- Test on your own documents. Run fifty of your real PDFs before you trust anyone’s benchmark, including this one.
- Keep a human on the numbers. One error in thirteen is fine for a draft and not fine for an invoice.
- Check availability in your own account. Launch posts are global. European rollouts often aren’t.
The stars are for navigation. The sun is for getting work done. For most of Europe’s small businesses, this autumn’s AI budget just moved into daylight.
Sources
Leave a Reply