dreamerv3 and dreamerv2

DreamerV3 is the successor to DreamerV2, extending discrete world models to diverse continuous control domains beyond Atari, making V2 largely superseded for new projects though both remain available implementations of the same algorithmic lineage.

dreamerv3
66
Established
dreamerv2
65
Established
Maintenance 2/25
Adoption 16/25
Maturity 25/25
Community 23/25
Maintenance 0/25
Adoption 15/25
Maturity 25/25
Community 25/25
Stars: 2,917
Forks: 484
Downloads: 412
Commits (30d): 0
Language: Python
License: MIT
Stars: 1,012
Forks: 210
Downloads: 238
Commits (30d): 0
Language: Python
License: MIT
Stale 6m No Dependents
Stale 6m No Dependents

About dreamerv3

danijar/dreamerv3

Mastering Diverse Domains through World Models

Builds a latent world model using categorical representations that predicts future states and rewards, then trains an actor-critic policy through imagined rollouts. Implemented in JAX with support for diverse environments (Atari, robotics, visual control tasks) and scales efficiently—larger models improve both final performance and sample efficiency without hyperparameter tuning across domains.

About dreamerv2

danijar/dreamerv2

Mastering Atari with Discrete World Models

Related comparisons

Scores updated daily from GitHub, PyPI, and npm data. How scores work