PetriBench: Benchmarking LLM Reasoning over Dynamic State Spaces
Pyrros Koussios and colleagues introduce PetriBench, which evaluates LLM reasoning over dynamic state spaces using Petri nets, a formalism for concurrent and distributed systems, with exact ground truth and procedural generation.
Ask this paper
Four task families by scope and temporal horizon. Each with Easy, Medium and Hard levels generated by increasing structural complexity, so difficulty is a controlled variable rather than a label.
Self-contained and extensible. No external knowledge is required and instances are generated rather than collected, which avoids the contamination and cost problems of hand-built reasoning benchmarks.
Hard instances separate models. Accuracy falls consistently with difficulty, and the harder levels expose distinct task-specific capability profiles across models.
Test-time compute interacts unevenly. More compute helps, but differently per reasoning task, which is a finding the benchmark's smooth structural scaling makes measurable.
Abstract
Characterizing LLM reasoning remains an open challenge, as many existing benchmarks isolate specific reasoning skills, rely on external knowledge, or are costly to extend. We introduce PetriBench, a compact, fully self-contained, and scalable benchmark for evaluating LLM reasoning over dynamic state spaces using Petri nets, a mature formalism for modeling real-world concurrent and distributed systems. PetriBench organizes reasoning into four task families varying by scope and temporal horizon, with Easy, Medium, and Hard levels generated by increasing structural complexity and evaluated against exact ground truth. Across a diverse set of proprietary and open-weight models, accuracy decreases consistently with difficulty, while harder instances expose increasingly distinct task-specific capability profiles. Additional analyses show that test-time compute improves performance but interacts differently with different reasoning tasks, and that procedural generation yields smooth scaling with structural complexity. Together, these results show that PetriBench provides a unified and extensible setting for probing the strengths, limits, and scaling behavior of LLM reasoning.