# Schematron V2 Turbo

> Schematron V2 Turbo is inference.net's 3B extraction model that turns messy HTML into clean JSON that conforms to a schema you supply. It offers the family's throughput- and cost-optimized option, faster than V2 Small while still beating the first-generation 3B model. 128K-token context, strict JSON schema output.

**ID**: `inference-net/schematron-v2-turbo`  
**Creator**: inference-net  
**Category**: text  
**Context**: 128K tokens  
**Pricing**: $0.03 / $0.15 per 1M · billed by your provider (BYOK)  
**Released**: 2026-04-16  
**Web page**: https://anyrouter.dev/model/inference-net/schematron-v2-turbo

**Input modalities**: text  
**Output modalities**: text  

**Capabilities**: chat, streaming, structured-output

## Usage

**Endpoint**: `POST https://anyrouter.dev/api/v1/chat/completions`

```bash
curl https://anyrouter.dev/api/v1/chat/completions \
  -H "Authorization: Bearer $ANYROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "inference-net/schematron-v2-turbo",
  "messages": [
    {
      "role": "user",
      "content": "Say hi in 3 words."
    }
  ]
}'
```

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| inference-net-byok | $0.03/M (list) | $0.15/M (list) |

BYOK routes are billed by your own provider key — AnyRouter charges $0. Rates marked (list) are the provider's published list price.

## Model card

- Total parameters: 3B [2]
- Native context: 128000 tokens [1]

### Strengths [1]

- Optimized for throughput and cost; processes pages faster than Schematron V2 Small.
- Strict JSON mode with full schema adherence.

### Known limits [1]

- Takes no system or user instructions; the JSON schema alone drives extraction, so ambiguous fields need explicit descriptions in the schema.
- Pages longer than the 128K context must be chunked or truncated.

### Use cases [1]

- High-volume HTML-to-JSON extraction where throughput and cost matter most.

### API notes [1]

- Send raw HTML as the user message and set response_format to json_schema with strict true.
- Keep temperature at 0; the model is trained to perform best there.
- Clean HTML with lxml first (drop scripts, styles and boilerplate) to match the training data.

### Links

- announcement: https://inference.net/blog/schematron
- docs: https://docs.inference.net/workhorse-models/schematron
- pricing: https://inference.net/models/schematron-v2-turbo/

### Sources

1. [Schematron — inference.net docs](https://docs.inference.net/workhorse-models/schematron) (accessed 2026-10-27)
2. [Schematron V2 Turbo — inference.net](https://inference.net/models/schematron-v2-turbo/) (accessed 2026-10-27)

## Source

- Upstream docs: https://docs.inference.net/workhorse-models/schematron