September 2025 Summaries

Schematron is a family of specialized models designed to convert messy HTML into structured JSON efficiently and affordably. Schematron-8B and Schematron-3B offer high extraction quality at a fraction of the cost and speed of traditional large language models (LLMs), making tasks like web scraping and data parsing economically feasible for large-scale applications. These models cater to varying complexities and context lengths, with Schematron-8B handling complex extractions and Schematron-3B excelling in simpler tasks, both guaranteeing parseable, schema-compliant output. The development of Schematron addresses the challenges of structured web data extraction by providing a cost-effective solution without compromising on accuracy, enabling new use cases such as real-time monitoring and large-scale internet scraping. Schematron models are made available through open source platforms like Hugging Face and a serverless API, encouraging developers to integrate them into their applications for extracting clean and structured data.