GPT-6.1 Sol
Made by OpenAI
Good on harder writing and planning, for less than OpenAI's top model costs.
Try it for: proposals, policies, tidy write-ups.
Start with: a proposal you already know how to judge.
Maker's model page ↗Updated 25 October 2026
The AI models I’d use for business
Start with the job you need done. Then choose a model, give it a fair test, and keep a person responsible for the result.
Checked 25 October 2026
Writing & planning
Big documents & projects
Research & numbers
Busywork at volume
Four starting points
A shortlist to test against your own work.
Claude Opus 5.5
Made by Anthropic
Built for extended projects and professional work.
Try it for: long reports, research write-ups, planning.
Start with: a long report with a clear brief.
Maker's model page ↗Gemini 4 Argon
Made by Google
Built for complex reasoning, including maths and science.
Try it for: pricing sums, checking data, technical questions.
Our own example: none published yet.
Maker's model page ↗GPT-6 Luna
Made by OpenAI
OpenAI's most efficient model, for lots of small, repeat jobs.
Try it for: sorting, tagging, pulling details out of forms.
Our own example: none published yet.
Maker's model page ↗Based on the makers' model documentation, checked 25 October 2026. Suggested jobs are editorial starting points, not comparative test results. Sources and how to choose →
A model is the engine. The app is the vehicle.
ChatGPT, Claude and Gemini are products you can open and use. The model is what does the work inside them. A product’s name does not tell you which model, tools or limits you get on your account. Check those before paying.
Try one real job before changing your week
- Choose something you can judge. Use a past quote, a short policy or a set of meeting notes. Remove customer details for the first test.
- Give each model the same brief. Include the facts, the intended reader, the format and what a useful result looks like.
- Count the work left for you. A fast draft that needs half an hour of repair may be worse than a slower, usable one.
- Check the facts yourself. Especially dates, calculations, product details and commitments to customers. An assured answer can still be wrong.
- Record the result. What did it produce, how long did review take, and what did the whole job cost?
When to try something else
If a small repeat job costs too much, test a more economical model. If a complex brief keeps falling apart, try a more capable one. Change one thing at a time so you know what helped. Better instructions, cleaner source material or a smaller job may matter more than a different model.
These are suggested starting jobs, inferred from the makers’ descriptions. They are not a measured ranking or a claim that I've personally tested every model. There is no universal winner here.
Sources and freshness
Model names and maker descriptions checked on 25 October 2026. Prices, access and features change, so use the linked documentation for the current details. We do not compare subscription prices with API token prices: they pay for different things.
Want a hand putting AI to work?
That's what we do at Beanpot. We run most of our own business on AI.