You don't need the most expensive AI. You need the right one for the job.
Writing an email? Solving a maths problem? Building software? Different AI models have different strengths. We'll help you find one that suits your task, and show a price wherever we've been able to check one.
Find the right AI See how the models compare
It takes about a minute to get started. Or just tell me.
Comparing 77 models using public benchmark data. Every published score links to its source.
No sponsored rankings. No invented scores. No tracking.
What do you need AI to do?
You don't need to know model names or technical jargon. Start with the job you want to get done.
- Write somethingEmails, articles, posts and ideas.Claude Opus 5.5 Best-supported now
- Code somethingBuild, debug or understand software.Claude Opus 5.5 Best-supported now
- Work through mathsCalculations, difficult problems and learning.GPT-6.1 Sol Best-supported now
- Get work doneBusiness emails, documents and everyday tasks.Gemini 4 Argon Best-supported now
- Research a questionExplore a subject and check factual claims.Gemini 4 Argon Best-supported now
Some models lead more than one job here: Claude Opus 5.5 for write something and code something; Gemini 4 Argon for get work done and research a question. That is not a mistake. Each pick comes from different tests, and the same model can come out first in more than one. How each pick is decided.
Meet the models
Different models are better at different things. Here's how the evidence compares.
Or browse by maker
- Alibaba 8
- Amazon 1
- Anthropic 12
- Baidu 2
- ByteDance 1
- DeepSeek 4
- Google DeepMind 11
- Meta 5
- Mistral AI 1
- Moonshot 3
- NVIDIA 1
- OpenAI 11
- Thinking Machines 1
- xAI 7
- Xiaomi 3
- Z.ai (Zhipu AI) 6
- Open weights 20
77 models are listed, chosen by a published rule rather than by hand (the top 60 on Arena's text vote, plus any model with at least three Epoch results). How models get included. Ratings compare only these models.
More capable doesn't always mean more useful.
A powerful AI can be worth paying for. But if you mainly write emails, summarise documents or brainstorm ideas, you may not need the most expensive option.
The sensible choice is the one that does your work well enough, consistently, at a price that makes sense for you.
On prices, we are careful. We show one only when we have seen it on the maker's own page, which is 8 of 77 models so far. For the rest we say “Price not independently verified” rather than guess. How prices work here.
We'll show you what the tests support, what they don't, and where the evidence runs out.
Recommendations you can check.
We don't ask you to take our word for it. Our comparisons use published benchmark data, and every displayed result links back to its source.
When a model hasn't been tested, we'll say so. When a test can't answer your question, we'll explain that too.
Your next AI doesn't have to cost more.
Tell us what you need help with. We'll show you what the evidence says.