- Job first, brand second. There is no single best AI, just as there is no single best vehicle.
- Start with the middle model. Move up only if the answers are not good enough.
- Test on your real task. That beats any ranking.
Step 1: name the job in one sentence
“I want to tidy up my emails.” “I want a poster with readable text.” “I want a short clip from a photo.” That sentence does most of the choosing.
| Write, ask, summarise: any of the big chat apps. The middle model is usually plenty. |
| Long documents: a model with a large context window (how much text it keeps in mind at once). |
| Images, video, voice: a tool built for that job. Text on posters, photo edits and video clips are different strengths. |
| Private work: a model you download and run yourself, or a business plan with a written privacy agreement. |
| Spend nothing: a chat app’s free plan, or a free local model if your computer is strong enough. |
Step 2: pick a step on the ladder
Most companies sell a strong, slower model for hard work, a middle one for everyday jobs, and a small, fast one for simple, repetitive work. Anthropic, for example, describes its top model as for “complex work requiring careful judgment”. 1
Step 3: app or download?
- An app sends your words to the company’s computers. Where they are stored depends on the company.
- A downloaded model keeps your chats on your machine, if your computer can run it. 2
- Every download has a licence. Some forbid business use. Read it before you sell anything you make.
Step 4: four checks before you commit
- Test it: same job, same instructions, two tools, side by side.
- Date any ranking: leaderboards change and disagree.
- Check it is not retiring soon: see when AI models retire.
- Find the training and memory settings before you paste anything private. Paying does not automatically switch training off. 3
The best model does your job well enough, at a price you accept, with privacy terms you are comfortable with. It can be a different one for each job.
- 1Anthropic, Claude Sonnet 5.5, 28 Sep 2026, read at source 1 Oct 2026Opus 5.5 handles “complex work requiring careful judgment”; Sonnet 5.5 is for “well-scoped everyday tasks”. ↑
- 2Ollama, FAQ, read at source 22 Sep 2026“Ollama runs locally. We don’t see your prompts or data when you run locally.” ↑
- 3Anthropic, Updates to Consumer Terms and Privacy Policy, 28 Aug 2025, read at source 23 Sep 2026“We will train new models using data from Free, Pro, and Max accounts when this setting is on”. ↑
The comparisons on this page (vehicles, the ladder) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK