AI article
StateBench: Testing how language models track state changes over long documents
huggingface | Nov 3, 2026 | Kyle Wiggers
Automated excerpt
StateBench consists of four difficulty levels that permit increasingly complex state update operations. General finite problems add assignment by reference. All four performed well on r-trivial and aperiodic problems.
Selected automatically from source text; not independently written or fact-checked. Read the original for full context.