ReaderLM v2: Small Language Model for HTML to Markdown and JSON

**What the feature is**
ReaderLM-v2 is a specialized 1.5B parameter small language model (SLM) designed specifically for HTML-to-Markdown conversion and structured JSON extraction. It supports documents up to 512K tokens across 29 languages and is optimized to strip clutter while preserving the semantic structure of complex web pages.

**Value proposition**
- Token Efficiency: Reduces LLM costs and latency by converting "noisy" HTML into clean, high-density Markdown, significantly lowering the input token count for downstream RAG systems.
- Structural Accuracy: Offers 20% higher accuracy than previous versions in handling complex web layouts, ensuring that headers, tables, and lists remain intact for better reasoning.
- Privacy & Speed: As a 1.5B parameter model, it is small enough to be deployed locally or on the edge, providing near-instant parsing without sending sensitive raw data to external providers.