Parameter tables rarely tell you which of two models suits a real conversation. lmarena.ai runs anonymous battles: one question goes to two unnamed models, you vote on the answers, identities are revealed afterward, and votes aggregate into leaderboards. It is open to use; entries and project splits follow the site page. It reflects user preference and shifts as models change, so treat it as a reference and test on your own tasks.
What it does
- Anonymous two-model comparison that limits brand bias
- Votes aggregate into capability leaderboards
- Projects across chat, code and other categories
- Rankings shift with model updates, not fixed verdicts
- Try the battle flow directly in the browser
Who it is for
- Building intuition about models during selection
- Comparing natural phrasing between two candidates
- Tracking tier changes on a task type
- Understanding evaluation method rather than marketing numbers
How to get started
- Use real questions from your own work, not small talk
- Repeat the same set several times to reduce one-off luck
- Note the candidates, then run your own task-level test
