About these results

How this was measured

These rankings reflect votes from 1,031 participating judges across the included boards and criteria. Judges compared outputs generated from an identical prompt set under identical settings. Judges did not see which model produced which output. Scores are computed using a Bradley–Terry model. Published results are current as of Sep 15, 2026 and correspond to OpenArt Arena index v1.0. Model versions are listed with each entry; providers update models without notice, and results apply only to the model versions, index version, and date shown.

What the numbers do and don't mean

These results reflect subjective human preference on a specific set of prompts. They are not a measure of overall model quality, safety, licensing suitability, or fitness for any particular use, and they may not predict results on your own prompts. Differences between models whose confidence intervals overlap are not statistically meaningful.

Money

No model provider paid for inclusion, placement, ranking, or favorable treatment in OpenArt Arena, and no provider had any input into the methodology, the prompt set, or the scoring. OpenArt has ordinary commercial relationships - such as licensing and API agreements - with some providers whose models appear here, and those relationships had no role in the results. We do not currently accept payment from model providers in connection with OpenArt Arena. If that changes, we will say so on this page and date the change.

Judges

Compensation varies by judge. Some receive a fee for completing a review task, some receive OpenArt platform credits, and some participate without compensation. In every case compensation is for participation only and is never conditioned on rating any model favorably. Judges affiliated with a model provider are excluded from that provider's comparisons.

• Affiliated raters. 21 of our 1,031 participating judges are OpenArt employees, contractors, or OpenArt Studios personnel, accounting for 1.82% of all ratings included in these results. The full-panel figures are published at https://openart.ai/arena/leaderboard.

Data

We do not use ratings, prompts, or generated images from OpenArt Arena to train, fine-tune, or build preference or reward models, and we do not license this data to others for that purpose. We use it to compute and publish these results, to improve the Arena itself, and for research we publish in aggregate. We do not share judge-level data - individual votes, comments, or identities - with model providers. Providers receive the same public aggregate results as everyone else, at the same time.

Who we are

OpenArt Arena is operated by OpenArt AI and covers models available on the OpenArt platform. It is not an independent or third-party certification, and it is not a comprehensive survey of the field.

Corrections

Model providers and readers who believe a result is inaccurate may write to arena@openart.ai. The full methodology is at Methodology. All model and company names are trademarks of their respective owners and are used for identification only; no affiliation or endorsement is implied.

License

The OpenArt Arena results are made available under a Creative Commons Attribution-NoDerivatives 4.0 International License.