New: the Timeline is live. Track world model releases, papers, and benchmark updates in real time.
world-models.io
The Knowledge Hub for AI World Models

World Model Benchmarks and Scores by Family

Benchmark evidence is organized by compatible tasks, protocols, and output types rather than a single cross-family score.

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, compare systems on the leaderboard, or read the Performance Index methodology.

Versioned benchmark framework

Results are only shown within a shared task, protocol, metric, and unit. Results reported by authors are not presented as reproduced runs.

Download the machine-readable benchmark dataset (JSON)

  • Driving Simulation World Models — 5 models, 0 shared protocols, 0 reported results. Protocol status: draft. Publish a scenario suite, sensor schema, simulator version, and safety metrics before accepting results.
  • Interactive Video World Models — 34 models, 0 shared protocols, 0 reported results. Protocol status: draft. Specify tasks, action space, latency measurement, rollout horizon, and evaluation capture procedure.
  • Spatial 3D World Models — 2 models, 0 shared protocols, 0 reported results. Protocol status: draft. Fix scene inputs, navigation tasks, reconstruction metrics, and export validation before accepting results.
  • Model-Based RL — 11 models, 2 shared protocols, 4 reported results. Protocol status: public. Run the open-model harness with recorded seeds, environments, hardware, checkpoints, and raw logs.
  • Latent Predictive World Models — 6 models, 0 shared protocols, 0 reported results. Protocol status: draft. Define downstream tasks, frozen-versus-finetuned settings, data splits, and transfer metrics.
  • Embodied Robotics World Models — 5 models, 0 shared protocols, 0 reported results. Protocol status: draft. Fix robot embodiment, task set, observation mode, simulator or hardware version, and safety reporting.