---
title: "DreamerV2 | First Human-Level Model-Based Atari Agent."
canonical: "https://world-models.io/en/models/dreamer-v2/"
description: "DreamerV2 was the first model-based agent to achieve human-level performance on Atari using discrete world model representations."
author: "Tyler D."
generated: "2026-10-30"
---

New: the Timeline is live. Track world model releases, papers, and benchmark updates in real time.

The Knowledge Hub for AI World Models

# DreamerV2

DreamerV2 was the first model-based reinforcement learning agent to achieve human-level Atari performance, introducing discrete categorical representations to the RSSM architecture.

robotics

model-based-rl

simulation

embodied-ai

## What is a world model?

A world model is an AI system that learns a predictive representation of how an environment changes over time. It can use observations, actions, or language to estimate future states and possible outcomes before an agent acts. World models support planning in model-based reinforcement learning, robotics, embodied AI, autonomous driving, and interactive simulation. This site catalogs these systems with model records, research papers, benchmark context, and source links. Start with the [world models database](https://world-models.io/en/world-models-database/), compare systems on the [leaderboard](https://world-models.io/en/leaderboard/), or read the [Performance Index methodology](https://world-models.io/en/methodology/performance-index/).

## Key Attributes

Static HTML snapshot of the model record for crawlers and no-JS readers.

| Attribute | Value |
| --- | --- |
| Model | DreamerV2 |
| Lab / Organization | DeepMind |
| Category | Model-Based RL |
| Subtype | Latent Dynamics Model |
| World Model Type | Imagination-based dynamics model |
| Primary Domain | Atari / Control |
| Architecture | RSSM with categorical discrete latent variables |
| Modality | Visual |
| Training Method | World model learning with discrete latent representations + imagination-based RL |
| Status | foundational |
| Year | 2021 |
| Performance Index | 72/100 (high confidence, v1.1) |

## About DreamerV2

Main editorial body preserved directly in static HTML.

DreamerV2 introduced discrete latent representations to world model learning, achieving human-level performance on the Atari 200M benchmark, a first for model-based methods. It demonstrated that discrete categorical variables can more effectively capture the multi-modal structure of complex environments than continuous representations.

DreamerV2 is an imagination-based dynamics model developed by Google in 2021 for atari / control.

## Editorial Snapshot

Short extractable facts for answer engines and no-JS readers.

| Signal | Value |
| --- | --- |
| Definition | DreamerV2 is an imagination-based dynamics model developed by Google in 2021 for atari / control. |
| Short Description | The first model-based agent to achieve human-level performance on the Atari benchmark using discrete world model representations. |
| Benchmark Rows | 1 |
| FAQ Entries | 1 |
| Related Models | 3 |
| Related Guides | 1 |
| Related Research Topics | 0 |
| Last Updated | 2026-02-18 |

## Notable Features

Key capabilities associated with this model.

- • First model-based method to reach human-level Atari
- • Discrete categorical latent representations
- • KL balancing for stable training

## Use Cases

Representative applications attached to this model record.

Atari games

Continuous control

Sample-efficient learning

## Strengths and Limitations

Balanced assessment surfaced in static HTML.

### Strengths

- • Human-level Atari
- • Discrete representations
- • More robust than V1
- • Sample efficient

### Limitations

- • Per-domain hyperparameter tuning still needed
- • Superseded by DreamerV3

## Benchmarks

Published benchmark evidence attached to this model record.

| Benchmark | Metric | Result | Source |
| --- | --- | --- | --- |
| Atari 200M | Mean HNS 1 x human | Human-level (first model-based) | [Source](https://arxiv.org/abs/2010.02193) |

## References and Citations

Primary references preserved in static HTML for citation extraction.

| Reference | Link |
| --- | --- |
| Hafner et al., 2021. Mastering Atari with Discrete World Models. ICLR 2021. | [Open source](https://arxiv.org/abs/1803.10122) |

## Related Models

Nearby models linked from the current editorial record.

| Model | Category | World Model Type | Index v1.1 |
| --- | --- | --- | --- |
| [DreamerV3](https://world-models.io/en/models/dreamer-v3/) | Model-Based RL | Imagination-based dynamics model | 88/100 |
| [PlaNet](https://world-models.io/en/models/planet/) | Model-Based RL | Latent space planning model | 57/100 |
| [RSSM](https://world-models.io/en/models/rssm/) | Latent Dynamics | Core dynamics architecture | 64/100 |

## Direct Comparisons

Side-by-side comparisons already connected to this model.

| Comparison | Matchup | Summary |
| --- | --- | --- |
| [DreamerV3 vs PlaNet](https://world-models.io/en/compare/dreamer-v3-vs-planet/) | DreamerV3 vs PlaNet | DreamerV3 represents the evolution of PlaNet's core ideas. Both use the RSSM architecture, but DreamerV3 adds discrete representations, symlog predictions, and fixed hyperparameters to achieve state-of-the-art performance across diverse domains. |
| [DreamerV2 vs PlaNet](https://world-models.io/en/compare/dreamer-v2-vs-planet/) | DreamerV2 vs PlaNet | Both use the RSSM architecture for latent dynamics, but DreamerV2 introduced discrete representations that dramatically improved performance. PlaNet pioneered the approach; DreamerV2 perfected it for Atari-scale environments. |
| [DreamerV3 vs IRIS](https://world-models.io/en/compare/dreamer-v3-vs-iris/) | DreamerV3 vs IRIS | Two model-based RL agents using fundamentally different world model architectures: DreamerV3's RSSM with actor-critic vs. IRIS's autoregressive Transformer with VQ-VAE tokens. |
| [Ha & Schmidhuber World Model vs DreamerV3](https://world-models.io/en/compare/ha-world-model-vs-dreamer-v3/) | Ha & Schmidhuber World Model vs DreamerV3 | The original 2018 'World Models' paper vs. the current state-of-the-art: how five years of research transformed a foundational concept into a domain-general world model agent. |
| [MuZero vs DreamerV3](https://world-models.io/en/compare/muzero-vs-dreamer-v3/) | MuZero vs DreamerV3 | Two titans of model-based RL with fundamentally different approaches: MuZero learns a value-equivalent model for search-based planning, while DreamerV3 learns a generative world model for imagination-based policy optimization. |
| [DreamerV2 vs DreamerV3](https://world-models.io/en/compare/dreamer-v2-vs-dreamer-v3/) | DreamerV2 vs DreamerV3 | The Dreamer lineage's two most impactful iterations: DreamerV2 achieved human-level Atari with discrete representations, while DreamerV3 eliminated hyperparameter tuning entirely with symlog predictions. |

## Guides Referencing This Model

Crawler-readable guide links tied to this model.

| Guide | Summary |
| --- | --- |
| [RSSM Architecture Explained: Dreamer World Model](https://world-models.io/en/guides/understanding-rssm/) | A technical guide to the Recurrent State-Space Model: the foundational architecture behind PlaNet and the Dreamer family of world models. |

## Frequently Asked Questions

FAQ answers rendered directly into static HTML for extractable responses.

### Why use discrete latent variables?

Discrete representations enable more expressive and robust world models that better capture distinct environment states and multi-modal dynamics.

## Quick Answer

Short extractable summary preserved directly in static HTML.

- DreamerV2 is an imagination-based dynamics model developed by Google in 2021 for atari / control.
- Use this page when you need a fast read on how DreamerV2 fits into the model-based rl landscape, then validate the details in the benchmarks, citations, and related pages.
- A key strength surfaced in the editorial record is human-level Atari.

## Editorial Trust Signals

Editorial provenance and refresh policy preserved directly in static HTML.

**Published by** world-models.io editorial board.

**Lead editor** Tyler D. - Technical editor, methodology and benchmark analysis.

This model page synthesizes primary papers, official model pages, benchmark evidence, and related world-models.io context into a reference resource.

Each editorial page is assembled from primary sources, normalized into extractable summaries, checked for factual drift, and reviewed before publication or major refreshes. Last reviewed: 2026-02-18.

Pages are refreshed when a new paper, benchmark, release, architecture update, or stronger primary source materially changes the answer a reader or AI system should retrieve.

Each page links back to relevant primary sources and keeps a stable canonical URL so readers can verify claims, trace context, and reference the most up-to-date version. [See the editorial policy.](https://world-models.io/en/editorial-policy/)

Primary sources only

Last reviewed date visible

Methodology documented

Source links included

Correction channel published

Review date visible

## External Sources

Primary model and lab sources embedded in static HTML.

- [DreamerV2 paper](https://arxiv.org/abs/2010.02193) Google primary research source
- [DreamerV2 cited source](https://arxiv.org/abs/1803.10122) Hafner et al., 2021. Mastering Atari with Discrete World Models. ICLR 2021.
- [Google DeepMind official website](https://deepmind.google) Related lab

## References

1. [[1] Hafner et al., 2021. Mastering Atari with Discrete World Models. ICLR 2021.](https://arxiv.org/abs/1803.10122)