WorldArena Challenge @ CVPR 2026

Important Dates

2026.03.06

Submission Opens

Challenge submission portal opens for participating teams.

2026.04.25

Mid-competition Ranking Release

The ranking results of mid-stage checkpoints will be released.

2026.05.15

Leaderboard Update

The leaderboard will be updated on this fixed date. For the remaining competition period, no leaderboard updates will be made outside May 15 and May 25.

2026.05.25

Final Submission Deadline & Leaderboard Update

Final date to submit results for Track 1 and Track 2 evaluation, and the second and final fixed leaderboard update date.

2026.06.04

Challenge Session & Awards at the CVPR 2026

Workshop challenge session and award announcements at CVPR 2026.

Challenge Overview

The WorldArena Challenge is the CVPR 2026 world model challenge built on WorldArena, a unified benchmark for evaluating embodied world models beyond visual realism alone.

Our goal is to evaluate whether a world model is not only perceptually convincing, but also functionally useful for embodied decision-making, simulation, and control. The challenge therefore emphasizes both perception quality and functional utility in a single evaluation setting.

Participants compete on two complementary tracks inspired by the official benchmark and the poster overview:

Track 1: Video Perception Quality — evaluating visual quality, motion quality, content consistency, physics adherence, 3D accuracy, and controllability.
Track 2: Data Engine & Policy Evaluator — evaluating generated worlds as embodied data engines and policy evaluators for downstream tasks.

By combining the two tracks, the challenge provides a holistic benchmark that encourages progress toward world simulators that are realistic, controllable, and useful for real embodied AI applications.

Tracks & Submission

Track 1

Video Quality

Evaluate video perception quality across visual quality, motion quality, content consistency, physics adherence, 3D accuracy, and controllability.

Resources:

Guide

Dataset Leaderboard

click to view full image

Track Group QR

HOT: Open for submission

Final Submission Deadline:

2026.05.25

Track 2

Data Engine & Policy Evaluator

Evaluate whether generated worlds preserve decision-relevant dynamics and support embodied tasks as data engines and policy evaluators.

Resources:

Guide

Dataset Leaderboard

click to view full image

Track Group QR

NOTICE: Track-specific resources continue to be updated

Evaluation Tasks:

Track 2 leaderboard evaluation covers the Data Engine and Policy Evaluator tasks.

Close-source models: Please provide API access for evaluation.

Submission Limit: Each team may submit at most twice per day. If a team submits more than twice on the same day, the latest two submissions of that day will be used for evaluation.