AI model rankings

A higher score indicates stronger preference in Arena’s comparisons. It is not a percentage of correct answers or a certificate of accuracy, safety or suitability for a particular task. Close positions should be interpreted alongside their confidence intervals.

Arena also publishes a style-adjusted ranking. Its order can differ from the unadjusted category shown here. Model names and configuration labels are retained as published.

Updates and attribution

On the first visit, this page checks Arena’s latest published dataset. Each browser saves the validated results and the time of the successful check. Later visits reuse them for six hours. Once that interval has elapsed, opening or reloading the page requests the source again; leaving the page open does not trigger a scheduled check. If browser storage is unavailable, each visit checks again. A failed check retains the last validated results, or the dated reference table if no successful check is available, and is reported explicitly. No new check time is recorded for a failed request. All check times are shown in UTC.

Data: Arena / LMArena. Reused under Creative Commons Attribution 4.0 International (CC BY 4.0). Editorial presentation: selection of thirty entries, English explanatory text and rounding to one decimal place. The source data retain their CC BY 4.0 licence independently of this website’s general licence. No endorsement by Arena is implied.

Automatic updates retrieve public data from Hugging Face, the host of Arena’s dataset. View the data used by this table.