--- title: "DreamerV2 vs DreamerV3 | Comparison | world-models.io" canonical: "https://world-models.io/en/compare/dreamer-v2-vs-dreamer-v3/" description: "Comparing DreamerV2 and DreamerV3: how the Dreamer family evolved from human-level Atari to domain-general mastery." author: "Bernard Grenat" generated: "2026-10-23" --- New: the Timeline is live. Track world model releases, papers, and benchmark updates in real time. The Knowledge Hub for AI World Models # DreamerV2 vs DreamerV3 DreamerV2 vs DreamerV3 compares two generations of the Dreamer world model family, showing the evolution from first human-level Atari to domain-general mastery with fixed hyperparameters. 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/). ## Comparison Overview Main comparison summary preserved directly in static HTML. 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. ## Verdict Primary editorial conclusion preserved for non-JS crawlers and readers. DreamerV3 is the clear successor: it retains DreamerV2's discrete representation insight while solving the hyperparameter tuning problem that limited DreamerV2's generality. DreamerV2 remains important as the model that proved discrete latent spaces could reach human-level Atari, a breakthrough that DreamerV3 builds upon. ## Key Differences Extractable difference list generated from the comparison table. - **Key Innovation:** DreamerV2 - Discrete categorical latent space; DreamerV3 - Symlog predictions + fixed hyperparameters. - **Hyperparameters:** DreamerV2 - Per-domain tuning required; DreamerV3 - Single set across all domains. - **Atari Performance:** DreamerV2 - Human-level (first model-based); DreamerV3 - Atari 100K: 125% mean / 49% median human-normalized. - **Open-World:** DreamerV2 - Not evaluated; DreamerV3 - First to collect diamonds in Minecraft. - **Objective:** DreamerV2 - KL balancing; DreamerV3 - KL balancing + symlog normalization. ## When To Use Each Static decision guidance for no-JS readers. ### Choose DreamerV2 when... Choose DreamerV2 when its documented capabilities match your research or deployment requirements. ### Choose DreamerV3 when... Choose DreamerV3 when its documented capabilities match your research or deployment requirements. ## Comparison Table DreamerV3 is the clear successor: it retains DreamerV2's discrete representation insight while solving the hyperparameter tuning problem that limited DreamerV2's generality. DreamerV2 remains important as the model that proved discrete latent spaces could reach human-level Atari, a breakthrough that DreamerV3 builds upon. | Dimension | DreamerV2 | DreamerV3 | | --- | --- | --- | | Key Innovation | Discrete categorical latent space | Symlog predictions + fixed hyperparameters | | Hyperparameters | Per-domain tuning required | Single set across all domains | | Atari Performance | Human-level (first model-based) | Atari 100K: 125% mean / 49% median human-normalized | | Open-World | Not evaluated | First to collect diamonds in Minecraft | | Objective | KL balancing | KL balancing + symlog normalization | | Year | 2021 | 2023 | ## Performance Index Snapshot High-level scoring context for the models referenced in this comparison. | Model | Category | Index v1.1 | Confidence | | --- | --- | --- | --- | | DreamerV2 | Model-Based RL | 72/100 | high | | DreamerV3 | Model-Based RL | 88/100 | high | | PlaNet | Model-Based RL | 57/100 | high | | RSSM | Latent Dynamics | 64/100 | medium | ## Frequently Asked Questions FAQ answers rendered directly into static HTML for extractable responses. ### Should I use DreamerV2 or DreamerV3? DreamerV3 in almost all cases. Its fixed hyperparameters mean you don't need domain-specific tuning, and it matches or exceeds DreamerV2 on every benchmark. ### What's the main architectural difference? The core RSSM is similar, but DreamerV3 adds symlog predictions for scale-invariant learning, making it robust across environments with vastly different reward scales. ## Quick Answer Short extractable summary preserved directly in static HTML. - DreamerV2 vs DreamerV3: this page compares where each system is stronger instead of forcing a universal winner. - Use the verdict for the short answer, then validate the trade-offs in the table, evidence sources, and benchmark context. - Related models and source links help connect this comparison to the broader world models landscape. ## Editorial Trust Signals Editorial provenance and refresh policy preserved directly in static HTML. **Published by** world-models.io editorial board. **Lead editor** Bernard Grenat. This comparison page publishes a direct answer, explicit trade-offs, and source-backed evidence that can be validated against primary materials. 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: 2026-09-27. 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 papers and official sources for the models discussed on this comparison page. - [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. - [DreamerV3 paper](https://arxiv.org/abs/2301.04104) Google DeepMind primary research source - [PlaNet paper](https://arxiv.org/abs/1811.04551) Google primary research source