SO-Bench: A Structural Output Evaluation of Multimodal LLMs
This addresses the problem of evaluating and improving multimodal LLMs for real-world, agentic applications requiring schema compliance, though it is incremental as it builds on existing structured generation work.
The authors tackled the lack of benchmarks for evaluating multimodal LLMs on structured outputs by introducing SO-Bench, a comprehensive benchmark covering four visual domains with over 6.5K JSON schemas and 1.8K image-schema pairs, revealing persistent gaps in model performance and showing that training can improve structured output capability.
Multimodal large language models (MLLMs) are increasingly deployed in real-world, agentic settings where outputs must not only be correct, but also conform to predefined data schemas. Despite recent progress in structured generation in textual domain, there is still no benchmark that systematically evaluates schema-grounded information extraction and reasoning over visual inputs. In this work, we conduct a comprehensive study of visual structural output capabilities for MLLMs with our carefully designed SO-Bench benchmark. Covering four visual domains, including UI screens, natural images, documents, and charts, SO-Bench is built from over 6.5K diverse JSON schemas and 1.8K curated image-schema pairs with human-verified quality. Benchmarking experiments on open-sourced and frontier proprietary models reveal persistent gaps in predicting accurate, schema compliant outputs, highlighting the need for better multimodal structured reasoning. Beyond benchmarking, we further conduct training experiments to largely improve the model's structured output capability. We plan to make the benchmark available to the community.