---
title: LightOnOCR-3 on SIE | Superlinked
description: SIE, an open-source inference engine, serves LightOnOCR-3, an open document model that turns PDFs and scanned pages into Markdown with their tables intact.
canonical_url: https://superlinked.com/models/lightonai-lightonocr-3-4b
last_updated: 2026-11-09
---

# LightOnOCR-3

An open document model that turns PDFs and scanned pages into Markdown with their tables intact.

Model id: `lightonai/LightOnOCR-3-4B` · Endpoint: `extract` · [Hugging Face](https://huggingface.co/lightonai/LightOnOCR-3-4B)

Listed under [Doc to Markdown and OCR](/models#documents) on [the model index](/models.md).

| Accuracy | Time per request | Serverless price |
|---|---|---|
| – (olmOCR-Bench, checks passed · not measured) | – (Not timed one request at a time) | $0.14 in · $1.37 out per 1M tokens |

## On the task pages

- [Convert pages to Markdown 48% cheaper than Mistral OCR 4.1](/doc-to-markdown): 73.2% Document checks passed · 8,234 checks; $2.05 Cost per 1,000 pages

## Call it

On SIE Serverless:

```python
from urllib.request import urlopen
from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")

image_bytes = urlopen("https://superlinked.com/reference/doc-to-markdown/methomyl-page.png").read()
image = {"data": image_bytes, "format": "png"}
result = client.extract(
    "lightonai/LightOnOCR-3-4B",
    {"id": "methomyl-page", "images": [image]},
    options={"max_new_tokens": 12288, "temperature": 0.1, "top_p": 1},
)
print(result["entities"][0]["text"])
```

## Next to the other document models

| Model | olmOCR-Bench, checks passed | Time per request | Serverless price | Size |
|---|---|---|---|---|
| [LightOnOCR-3](/models/lightonai-lightonocr-3-4b.md) `lightonai/LightOnOCR-3-4B` | – | – | $0.14 in · $1.37 out per 1M tokens | 4.5B |
| [GLM-OCR](/models/zai-org-glm-ocr.md) `zai-org/GLM-OCR` | 73.8% | – | $0.25 per 1,000 images | 1.3B |
| [LightOnOCR 2 1B](/models/lightonai-lightonocr-2-1b.md) `lightonai/LightOnOCR-2-1B` | 77.7% | – | $2.00 per 1,000 pages | 1.0B |
| [MinerU2.5 Pro 2604 1.2B](/models/opendatalab-mineru2-5-pro-2604-1-2b.md) `opendatalab/MinerU2.5-Pro-2604-1.2B` | 59.1% | – | Not on Serverless (run in your cloud) | 1.2B |
| [PaddleOCR VL 1.5](/models/paddlepaddle-paddleocr-vl-1-5.md) `PaddlePaddle/PaddleOCR-VL-1.5` | 51.7% | – | Not on Serverless (run in your cloud) | 959M |

– = not measured / not timed / not listed.

[All 6 document models](/models#documents)

## Facts

- Size: 4.5B parameters
- License: Apache-2.0
- Reads: Image
- Endpoint: extract

## Evaluation

- Accuracy: Run on SIE against the public dataset named in each figure
- Prices: Serverless list prices
- Sources: [Task page runs and their inputs](https://huggingface.co/datasets/superlinked/sie-task-evidence)