Benchmark RadarAI BENCHMARK PROFILE
Chess-World-Model
Chess-World-Model evaluates state tracking by predicting the exact board state after sequences of legal moves from 10 million chess games, with accuracy as the metric.
- Released
- 2026-05-28
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
State tracking is under-tested in realistic domains. This benchmark exposes limitations in Transformer and RNN models that scale alone may hide, offering a practical testbed for world models.
Motivation
World models require state tracking, which is the ability to maintain a correct latent state across action sequences.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.