BENCHMARK GUIDE
What does an Arena score mean?
Arena asks people to compare anonymous model responses. Those preferences contribute to relative model rankings.
When it is useful
Use it to explore which responses people tend to prefer in the category you care about. Record the leaderboard category and snapshot date.
What it cannot tell you
Preference is different from independently verified correctness. A rating is not a percentage of tasks solved, and rankings can shift as votes and comparison models change.
A clearer, more readable answer may be preferred even when you still need to check its facts. For a business workflow, measure whether it completes the task as well as whether people like its style.
The chart shows selected published results with a source, snapshot date and tested configuration. Coverage differs by benchmark. Missing scores stay blank; demo values are no longer shown.
Sources
Developer reports describe their own results. This site has not independently reproduced them.
Arena: how evaluation works Chatbot Arena research paper