World Model Cognitive Benchmark for Embodied AI

WM Bench v1.0 is the first benchmark designed to evaluate the cognitive capabilities of World Models and Embodied AI systems. It was created by FINAL-Bench and released in March 2026. The benchmark moves beyond measuring visual fidelity to assess a model's reasoning and understanding.

Use Cases

• Benchmarking a world model's ability to predict future states based on its internal representations.
• Evaluating an embodied AI system's performance on tasks requiring planning and reasoning, as defined by the benchmark's metrics.
• Comparing different model architectures on their cognitive capabilities using the provided evaluation protocol.

Strengths

• First benchmark specifically for world model and embodied AI cognitive evaluation.
• Released in March 2026, indicating recent development.

Limitations

• Specific dataset size, row count, and column details are unknown.
• The full description and task specifics require visiting an external page.

Provenance

Source

FINAL-Bench

Freshness

Last updated 2026-03-29.