PaddleOCR-VL 1.5 - Free Document Parsing Model on SiliconFlow, 16K Context
PaddlePaddle/PaddleOCR-VL-1.5
Model Information
| Property | Value |
|---|---|
| Context length | 16,384 tokens |
| Max output | 16,384 tokens |
| Input price | ¥0 / 1M tokens |
| Output price | ¥0 / 1M tokens |
| Provider | Baidu PaddlePaddle / SiliconFlow |
| Status | Free |
| Input modalities | Text + image |
| Parameters | 0.9B |
| Release date | 2026-01-29 |
Description
PaddleOCR-VL-1.5 is a lightweight multimodal document parsing model from Baidu. With only 0.9B parameters, it reaches 94.5% accuracy on the OmniDocBench v1.5 benchmark, surpassing general-purpose multimodal models such as Gemini-3 Pro. The new version supports localization of irregularly shaped bounding boxes and performs reliably in real-world conditions such as scans, skewed or folded pages, screen captures, and complex lighting, while adding seal recognition together with text detection and recognition.
The model supports 111 languages and can automatically merge tables that span pages and recognize section headings, making it suitable for document digitization, invoice and contract parsing, and multilingual material organization.
Usage Example
curl https://api.siliconflow.cn/v1/chat/completions \
-H "Authorization: Bearer $SILICONFLOW_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "PaddlePaddle/PaddleOCR-VL-1.5",
"messages": [{
"role": "user",
"content": [
{"type": "text", "text": "Extract the text and table structure from this image"},
{"type": "image_url", "image_url": {"url": "https://example.com/document.png"}}
]
}]
}'Related Links
📚 Related Free Models
These models are also available for $0 on SiliconFlow. Choose by context length and modality.
| Model | Context | Description |
|---|---|---|
| Hunyuan-MT-7B | 33K | Hunyuan-MT-7B is Tencent Hunyuan’s lightweight 7B |
| Xing4.0 29B | 256K | TeleAI’s open-source 29B MoE model with 4B active |
📋 View all SiliconFlow free models →