ParseBench is the first benchmark to include VLM chart understanding 📊📈📉 over enterprise document...
TL;DR · AI 摘要
ParseBench是首个评估VLM在企业文档中理解嵌入图表能力的基准，包含568页真实文档和多种图表类型，填补现有基准仅关注独立图表的空白。
核心要点
- 现有基准如ChartQA仅测试独立图表，忽略图表在完整文档中的上下文语义。
- ParseBench提供568页含多样化嵌入图表的真实企业文档，支持离散/连续序列等复杂类型。
- 每个图表数据点由模型初筛+人工校验，确保评估AI代理提取数值能力的准确性。
🟠 Existing benchmarks (ChartQA, ChartXiv) test over charts specifically and not the chart's inclusion in the overall document. Also doesn't contain references to real-world https://t.co/yL8IoTDn7X" / X
ParseBench is the first benchmark to include VLM chart understanding 
over enterprise documents. 
Existing benchmarks (ChartQA, ChartXiv) test over charts specifically and not the chart's inclusion in the overall document. Also doesn't contain references to real-world docs 
ParseBench contains 568 pages containing a diversity of charts embedded in real-world documents. It contains a mix of charts: discrete series, continuous series, bar/point/line graphs, charts without clear markers, and more Each chart has a set of ground-truth datapoints bootstrapped with an initial model and verified through human annotators (with a tolerance) Come check it out! Blog: llamaindex.ai/blog/parsebenc Paper: arxiv.org/abs/2604.08538 Website: parsebench.ai/?utm_medium=so
Quote
LlamaIndex 
@llama_index
9h
Let's talk parsing charts . Last week we released ParseBench, the first document OCR benchmark for AI agents. New in ParseBench: ChartDataPointMatch. Most document look at a chart and OCR the caption. Agents need the actual numbers. That's the gap between "OCR'd the
