AGI Maze supplies a family of grid maze environments with a clean API to benchmark agents on learning and using world state representations rather than local pattern matching, with preliminary tests showing vanilla LLMs fail even on small instances.
beat the benchmark
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
AGI Maze as a Benchmark Framework for World-Modeling Agents
AGI Maze supplies a family of grid maze environments with a clean API to benchmark agents on learning and using world state representations rather than local pattern matching, with preliminary tests showing vanilla LLMs fail even on small instances.