Free tool

Which AI model should you use?

Answer five questions about what the AI should do. Get a suggested model size, why, and what each Claude, GPT and Gemini model in it would cost you a month.

Prices checked

12 October 2026, from each vendor's published pricing. Costs are planning estimates.

Also try the AI cost calculator and the token counter.

1.What should the AI do?
2.How good must the answers be?
3.How many requests a month?
4.How much text goes in each request?
5.Does someone wait for the answer?

Our suggestion

Small & fast model

  • · Answer customer questions is well within what small models do reliably.

Small & fast models, cheapest first for your workload

  • Claude Haiku 5.5 · Anthropic$6.50/mo
  • GPT-6 Luna · OpenAI$6.50/mo
  • Gemini 3.5 Flash-Lite · Google$22.50/mo

Assumes 10,000 requests, 5,000-token prompts, 300-token answers, no caching. Fine-tune the numbers in the AI cost calculator.

Want a second opinion on your real use?

Tell us what the AI does and we'll test the shortlist on your own examples, then tell you which model and setup to use.

Ask about my use case →
How it works

Start small, and step up only when the answers fall short.

Most teams pick the biggest model by default and pay several times more than they need to. The better habit is to test a smaller model on fifty real examples, and move up only where it fails.

Which AI model is best for a small business?
Usually the cheapest one that does the job well. Sorting, short answers and summaries rarely need a flagship model; writing, document analysis and multi-step agents often do. Test a smaller model on your real examples before paying for a bigger one.
Why does the picker suggest a size, not one brand?
We do not publish benchmarks, so we do not rank Claude, GPT and Gemini against each other. Within the suggested size, models are listed by what your workload would cost, using each vendor's published prices.
How do you decide the size?
From our experience building these systems: the task sets a starting size, then the quality you need, your volume and how much text each request carries move it up or down. The reasons are shown with every result.
How accurate are the costs?
They use the same published prices as our AI cost calculator and rough answer lengths per task. Real bills depend on your prompts and answers, so treat them as a planning range.
Paying more than you expected?
The AI Cost Review looks at your real usage and shows where the bill can come down without hurting the answers.

Building something with AI?

Tell us what it should do. We'll tell you which model fits, what it will really cost, and build it.