fiready/text-wallet-hotkey3

TEXT GENERATIONPricing:Input $0.2036 / Output $1.34Concurrent Unit Cost:1Model Size:3.2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 22, 2025License:llama3.2Architecture:Transformer Featherless Exclusive Cold

The fiready/text-wallet-hotkey3, part of Inference.net's Schematron series, is a 3.2 billion parameter model specialized in converting noisy HTML into clean, schema-conformant JSON. With a long context window of up to 128K tokens, it excels at robust data extraction from lengthy web pages. This model is purpose-trained for web scraping, data ingestion, and transforming arbitrary web content into structured records, ensuring 100% schema-conformant JSON outputs.

Loading preview...

Model Overview

fiready/text-wallet-hotkey3 is the 3.2 billion parameter version of Inference.net's Schematron series, designed for advanced HTML-to-JSON extraction. This model is specifically trained to convert raw or cleaned HTML into strictly valid JSON that adheres to a provided JSON Schema. It is optimized for web scraping, data ingestion, and structuring unstructured web content.

Key Capabilities

  • Schema-First Extraction: Guarantees 100% schema-conformant JSON outputs, eliminating the need for post-processing to fit a schema.
  • Long Context Window: Supports up to 128K tokens, making it robust for processing lengthy and noisy HTML documents.
  • High Extraction Quality: Achieves a strong LLM-as-Judge score of 4.41 for HTML-to-JSON extraction quality, outperforming general-purpose models like Gemini-3B-Base.
  • Factuality Improvement: When integrated into a web-augmented pipeline, Schematron significantly boosts LLM factuality, improving accuracy from 8.54% to 82.87% for tasks like SimpleQA by providing structured data from web searches.

Use Cases

  • Web Scraping: Efficiently extract structured data from websites based on predefined schemas.
  • Data Ingestion: Transform arbitrary web pages into clean, typed JSON records for databases or analytics.
  • LLM Augmentation: Enhance the factuality and reliability of large language models by providing them with structured, schema-validated data from web searches, reducing token usage compared to processing raw HTML.

Recommendations

For optimal results, it is recommended to use a temperature of 0 and JSON mode for deterministic outputs. Pre-cleaning HTML to remove scripts and styles is advised, and validating the model's JSON output against your schema is a good practice.

Finetunes
1 models