A new set of open-weight models is topping the leaderboard for document understanding 🔥 INF AI just released two models: Infinity-Parser2-Pro (35B) and Infinity-Parser2-Flash (2B) that top our Hugging Face leaderboard for ParseBench. Two key insights: ✅ An expanded synthetic data engine over 5 million diverse parsing samples ✅ A novel Joint RL algorithm that co-optimizes multiple complex tasks: document parsing, element parsing, chart parsing, and more. ParseBench is an open benchmark designed to test semantic document understanding over real-world enterprise documents; it has comprehensive metrics over tables, charts, semantic formatting, and more. Come check out the results on ParseBench! HuggingFace 🤗: https://lnkd.in/gaZGbH_a Site: https://www.parsebench.ai/ Infinity-Parser Flash model: https://lnkd.in/gr6qkBBD
Mark Schmeer
1w
Document understanding is quietly becoming core infrastructure for AI systems. Most enterprise knowledge still lives inside PDFs, tables, screenshots, and messy formatting.
Vinay Mehta
1w
How does ParseBench handle multi-column tables where the semantic meaning depends on row context? That's where most parsers I've tested fall apart.
Interesting direction. Most people see document parsing as an extraction problem. The deeper layer is that structure itself becomes intelligence. Once a system can reliably understand relationships between charts, formatting, semantic hierarchy, and context at scale, you move from “reading documents” to modeling decision architecture. The next leap in AI won’t come from larger outputs alone. It’ll come from systems that understand how information is organized before reasoning even begins.