NEWΔ-IRIS·ICML 2024

WMWM Arena: World Model Predictions vs Real Atari Gameplay Benchmark

What do World Models dream?

Side-by-side visual comparison of World Model predictions vs ground truth across Atari games. Evaluate how faithfully each model imagines the future.

22 World ModelsAtari 100k64x64 @ 15fps
Explore World ModelsTry Blind Battle →
Visual Comparison
Breakout
SSIM 0.9566DIAMOND
Ground Truth
Breakout real gameplay
DIAMOND Prediction
Breakout DIAMOND prediction
0.9566
SSIM ↑Structural Similarity Index. Measures pixel-level structural similarity. Range 0-1, higher = more similar.
8.6
FVD ↓Frechet Video Distance. Measures distribution-level similarity of video features. Lower = closer to real.
0.0201
LPIPS ↓Learned Perceptual Image Patch Similarity. Measures perceptual similarity via deep features. Range 0-1, lower = more similar.
100
Frames
View details →
Featured Games
Asterix0.8722
Real
Asterix real gameplay
Predicted
Asterix DIAMOND prediction
MsPacman0.9824
Real
MsPacman real gameplay
Predicted
MsPacman DIAMOND prediction
Pong0.9424
Real
Pong real gameplay
Predicted
Pong DIAMOND prediction
Qbert0.9885
Real
Qbert real gameplay
Predicted
Qbert DIAMOND prediction
Model Leaderboard
#
Model
Architecture
HNS
Venue
Status
1
EfficientZero V2Ye et al. · Paper · Code
MuZero + MCTS
2.22
ICML 2024 Spotlight
Paper Only
2
EfficientZeroYe et al. · Paper · Code
MuZero + Self-Supervised
1.94
NeurIPS 2021
Paper Only
3
EDELINELee, Lin, Sun, Lee · Paper · Code
Mamba SSM + Diffusion
1.87
NeurIPS 2025
Paper Only
4
M³Cohen et al. · Paper · Code
Modular Transformer
1.65
2025
Paper Only
5
TWISTERBurchi, Timofte · Paper · Code
Transformer + CPC
1.62
ICLR 2025
Evaluated
6
DIAMONDAlonso et al. · Paper · Code · Weights
Diffusion (EDM)
1.46
NeurIPS 2024 Spotlight
Evaluated
7
HarmonyDreamMa et al. · Paper · Code
RSSM + Harmony Loss
1.36
ICML 2024
Paper Only
8
EMERALDBurchi, Timofte · Paper · Code
MaskGIT + Spatial Latent
1.34
ICML 2025
Paper Only
9
STORMZhang et al. · Paper · Code
Stochastic Transformer
1.27
NeurIPS 2023
Evaluated
10
OC-STORMZhang et al. · Paper · Code
Object-Centric Transformer
1.25
2024
Training
11
REMCohen et al. · Paper · Code
RetNet + Parallel Observation
1.22
ICML 2024
Paper Only
12
GIT-STORMMicheli et al. · Paper
Transformer + MaskGIT Prior
1.13
ICLR 2025
Paper Only
13
DreamerV3Hafner et al. · Paper · Code
RSSM + Symlog
1.12
JMLR 2024
Paper Only
14
Δ-IRISNEWMicheli, Alonso, Fleuret · Paper · Code
Transformer + Delta Tokens
1.11
ICML 2024
Evaluated
15
DARTAgarwal, Andrews, Kahou · Paper · Code
Fully Discrete Tokens
1.07
ICML 2024
Paper Only
16
DramaWang et al. · Paper · Code
Mamba-2 SSM
1.05
ICLR 2025
Paper Only
17
IRISMicheli, Alonso, Fleuret · Paper · Code · Weights
Transformer + dVAE
1.05
ICLR 2023
Evaluated
18
TWMChen et al. · Paper · Code
Transformer + VQ-VAE
0.92
ICLR 2023
Training
19
BBFSchwarzer et al. · Paper · Code
Scaled ResNet DQN
~0.92
ICML 2023
Paper Only
20
SPRSchwarzer et al. · Paper · Code
Rainbow DQN + Self-Prediction
0.70
ICLR 2021
Paper Only
21
DreamerV2Hafner et al. · Paper · Code
RSSM + Discrete Latent
0.67
ICLR 2021
Paper Only
22
SimPLeKaiser et al. · Paper · Code
VAE + Video Prediction
0.44
ICLR 2020
Paper Only