compare answers from several AIs

STEP BY STEPFrom paying for reputation to choosing on evidence

1. 01

   Go to arena.ai

   There is nothing to install. Open arena.ai in your browser, on a computer or a phone. Arena began in 2023 at the University of California, Berkeley as a research project called Chatbot Arena. It is a company now and since January 2026 it is called simply Arena. People's votes build its public ranking of AI models.

2. 02

   Create your account with Google or your email

   Select Login. If you choose Google, you sign in with that account and follow the prompts; in my test it asked for nothing else. If you use your email, enter the address, choose Continue with email, and confirm the account from the message you receive. Use an everyday address: Arena may reject temporary ones. You can try it without an account, but with one you get a higher daily limit, reach your history from different devices, and can use Direct, Side by Side and the image and video arenas.

3. 03

   Choose Battle Mode

   The mode selector is at the top. Battle Mode is the one to start with: two models answer your question without showing their names, and your vote counts towards the ranking. Side by Side lets you pick the two models, but you already know which is which and the vote doesn't count. Direct is talking to a single model. Start in Battle Mode: the point is to vote without knowing who wrote each answer.

   prompt · the same question for two AIs

   Copy

   Write the WhatsApp message to chase a client who is [two weeks] late paying [what for: for example, the August invoice], without sounding rude. Three lines maximum. Tone: [warm and professional].

   Paste it into Battle Mode with your own details in the brackets. Always state the format and the length (three lines, a list of five points, a two-paragraph email). That way you compare two answers to the same task rather than two different answering styles.

4. 04

   Read both answers and vote

   Two answers appear, A and B, with no names. Read them as if you were about to send the message as written: which would you send to your client? Below there are four buttons: A is better, B is better, tie, or both are bad. Vote what you actually think. There is no correct answer. Vote for the one that would serve your business better.

5. 05

   Find out which was which

   After you vote, Arena reveals each model's name. That comparison is the heart of the guide: you find out whether the answer you preferred came from the AI you pay for, from a free one, or from one you had never heard of.

   IN MY TEST

   I picked the answer with two options and found DeepSeek had written it; the softer one was Gemini. I was paying for neither.

6. 06

   Repeat with three real questions from your business

   One test is not a comparison. Ask three questions you genuinely have to solve this week: a message to a client, a summary of a long document, an idea for a promotion. Write down which model won each. If one wins two out of three, you have a candidate. If a different one wins each time, you may not need to pay for the most expensive model.

7. 07

   Confirm your candidate in Side by Side

   Once you have two favourites, switch to Side by Side and choose them yourself. Ask them the same question again and compare without rushing. You are not voting blind here, you are deciding which to use. Direct lets you try the winner before paying for a plan.

8. 08

   Read the ranking with judgement

   Leaderboard shows the public table built from user votes. There are rankings by task type: text, web pages, images, video, search and more. Use it as orientation rather than verdict: the ranking measures what people prefer in general, and your business has specific questions. Your three tests are worth more to you than the number one spot.

NEXT SUPERPOWER