JSONSchemaBench: A Rigorous Benchmark of Structured Outputs for Language Models

https://arxiv.org/abs/2501.10868

10K real-world JSON schemas; evaluates Guidance, Outlines, llama.cpp, XGrammar, OpenAI, and Gemini on efficiency, constraint coverage, and output quality, plus the official JSON Schema Test Suite.

Grammar-constrained decoding for valid game actions