Final Score = (0.2 × Render Score) +
(0.1 × Keyword Matching) +
(0.7 × VQA Score)
StructEval
As Large Language Models (LLMs) become integral to software development workflows, their ability to generate structured outputs has become critically important. We introduce StructEval, a comprehensive benchmark for evaluating LLMs' capabilities in producing both non-renderable (JSON, YAML, CSV) and renderable (HTML, React, SVG) structured formats. Unlike prior benchmarks, StructEval systematically evaluates structural fidelity across diverse formats through two paradigms: (1) generation tasks, producing structured output from natural language prompts, and (2) conversion tasks, translating between structured formats.
Our benchmark encompasses 18 formats and 44 types of tasks, with novel metrics for format adherence and structural correctness. Results reveal significant performance gaps—even state-of-the-art models like o1-mini achieve only 75.58% average score, with open-source alternatives lagging approximately 10 points behind. We find generation tasks more challenging than conversion tasks, and producing correct visual content more difficult than generating text-only structures.
StructEval comprises 2,035 examples covering 44 unique structure generation tasks across 18 structured output formats. The dataset is organized into two main subsets:
Evaluates text-only structured outputs
Evaluates visually rendered outputs
Please output JSON code. Task: Summarize metadata about a fictional scientific article. Feature Requirements: 1. Top-level field "title" is a string 2. Field "authors" is a list of exactly two items 3. Each author has "name" and "affiliation" 4. Field "publication.year" is an integer 5. Field "keywords" is a list of strings
titleauthors[0].nameauthors[1].affiliationpublication.yearkeywords[2]Please output HTML code. Task: Design a webpage for a travel itinerary. Feature Requirements: • Centered <h1> with "Trip Summary" • Use a <table> with 3 rows and 2 columns • Apply class "highlight" to second row • Add <button> labeled "Export PDF"
Our evaluation framework employs four core metrics:
Binary metric (0 or 1) indicating whether the generated code can be successfully loaded or rendered without syntax errors
Verifies structural correctness (existence of required keys, relationships between keys, etc.) using dot-path rules. Calculated as the percentage of dot-path rules satisfied by the generated output format.
Evaluates presence of desired keywords using exact string matching. Calculated as the percentage of keywords found in the raw generated output code.
Assesses visual correctness of rendered content through question-answer pairs. Calculated as the percentage of Q&A pairs satisfied by the rendered output.
HTML, React, SVG, LaTeX, Mermaid, etc.
Final Score = (0.2 × Render Score) +
(0.1 × Keyword Matching) +
(0.7 × VQA Score)
JSON, XML, YAML, CSV, TOML
Final Score = (0.2 × Render Score) +
(0.8 × Syntax Score)
We evaluate various state-of-the-art LLMs in a zero-shot setting. The table below shows the performance breakdown across our four task categories. Click on the column headers to expand detailed results.
| Model | Type | StructEval-T | StructEval-V | Average | ||
|---|---|---|---|---|---|---|
| Generation | Conversion | Generation | Conversion | |||
Performance by Task Type
Challenging Formats (Average Score < 50%)
Even state-of-the-art models struggle with structured output generation. GPT-4o achieves only 76.02% average score, while the best open-source model (Qwen3-4B) lags at 67.04%.
Generation tasks are generally more challenging than conversion tasks. Visual rendering (StructEval-V) proves harder than text-only structures (StructEval-T).
Several formats remain particularly difficult for all models: Text→TOML (35.8%), Text→Mermaid (18.9%), and Matplotlib→TikZ (28.4%) conversions.
Some tasks are effectively solved with scores >90%: JSON, HTML, CSV generation and YAML→JSON, React→HTML conversions show near-perfect performance.
@misc{yang2025structeval,
title={StructEval: Benchmarking LLMs' Capabilities to Generate Structural Outputs},
author={Jialin Yang and Dongfu Jiang and Lipeng He and Sherman Siu and Yuxuan Zhang and Disen Liao and Zhuofeng Li and Huaye Zeng and Yiming Jia and Haozhe Wang and Benjamin Schneider and Chi Ruan and Wentao Ma and Zhiheng Lyu and Yifei Wang and Yi Lu and Quy Duc Do and Ziyan Jiang and Ping Nie and Wenhu Chen},
year={2025},
eprint={2505.20139},
archivePrefix={arXiv},
primaryClass={cs.SE},
doi={10.48550/arXiv.2505.20139}
}