Schematron V2 Small is a specialized HTML-to-JSON extraction model from Inference.net that turns messy HTML into clean, schema-conforming JSON. It is designed for structured extraction workflows where the schema, not a prompt, defines the output shape.
The model is described as a 3B-parameter model and is positioned as the small member of the Schematron V2 family, with extraction quality aimed at complex schemas and long pages.
| Developer | Inference.net |
|---|---|
| Model ID | inference-net/schematron-v2-small |
| Release Date | September 12, 2026 |
| Context Window | 128,000 tokens (~128K) |
| Max Output Tokens | 4,096 tokens |
| Modalities | Text → Text |
| Capabilities | streaming, structured (JSON schema) output |
Specialized extraction model
Schematron V2 Small is built specifically for HTML-to-JSON extraction rather than general chat or reasoning tasks.
Schema-driven output
Extraction instructions are supplied through JSON Schema, Pydantic, or Zod, and the model returns schema-conforming structured output.
Strict JSON behavior
Inference.net says the model uses schema-constrained decoding and strict JSON mode so the output is valid JSON by construction.
Long-context support
The model supports a 128K token context window.
Streaming and structured output
Provider catalog listings identify streaming and structured_output support for this exact model.
Web scraping
The model is targeted at web scraping workflows that need reliable extraction from raw HTML into typed JSON.
Product data ingestion
It is presented as useful for turning e-commerce or catalog pages into structured feeds.
Financial document parsing
Provider and Inference.net materials describe it as suitable for extracting structured data from financial documents and filings.
Search-augmented pipelines
It is described as fitting pipelines that need dependable structured output over long pages.
Schematron 8B
Inference.net states that requests sent to inference-net/schematron-8b are now automatically routed to inference-net/schematron-v2-small.
Sources: https://inference.net/blog/schematron-v2/ · https://developer.puter.com/ai/models/ · https://inference.net/models/schematron-v2-small/
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Schematron V2 Small is suited to web scraping and HTML-to-JSON extraction, where it turns messy HTML into clean, schema-conforming JSON. It is also documented for product data ingestion, using a user-defined schema to return structured output from raw HTML. Another stated use case is financial document parsing, alongside other workflows that need reliable structured output over long pages. Inference.net also describes it as a fit for converting arbitrary pages into structured data.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "inference-net/schematron-v2-small", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "inference-net/schematron-v2-small", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"inference-net/schematron-v2-small","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
Schematron V2 Small is an HTML-to-JSON extraction model from Inference.net that turns messy HTML into clean, schema-conforming JSON. It is built for structured extraction rather than general chat.
It takes raw HTML plus a JSON schema, such as Pydantic, Zod, or JSON Schema definitions. Extraction instructions are carried through the schema rather than through a system or user prompt.
It returns schema-conformant structured output in JSON. Inference.net describes the output as valid JSON by construction and strictly schema-compliant.
Yes, it supports structured output. The model is designed to return validated JSON that conforms to the provided schema.
Yes, the provider catalogue lists streaming as a capability for this model. The model is also described as having strict JSON mode for schema-compliant output.
It is targeted at web scraping, product data ingestion, financial document parsing, and search-augmented pipelines that need reliable structured output. Inference.net also describes it as useful for complex schemas and long pages.
Do not write a conversational prompt; send the schema and the HTML instead. The schema acts as the interface for extraction.
No. It is a specialized extraction model focused on HTML-to-JSON conversion and schema-constrained decoding.
It is intended for messy HTML and long pages that need clean typed JSON extraction. Inference.net says it delivers frontier-level extraction quality on complex schemas and long pages.
Developers usually look for whether they need reliable HTML-to-JSON extraction, schema-constrained output, and structured data pipelines. They also check that their workload fits a model built for extraction rather than open-ended generation.
