PaddlePaddle
PaddleOCR-VL 1.6
About
PaddleOCR-VL 1.6 is a 0.9B open-weight model from PaddlePaddle. PaddleOCR-VL-1.6 for OCR, tables, formulas, and charts. It accepts text, images, and documents. Context window is 16K. Released May 27, 2026.
Parameters
0.9B
Published parameter count.
Context window
16K
Tokens of context on a request.
License
Apache-2.0
License on the weights.
Pricing
Token rates are US dollars per 1M tokens. Image rates are per 1K images. Audio rates are per 1M audio seconds.
| Rate | Price |
|---|---|
| Input / 1M | $0.15 |
| Cached input / 1M | $0.15 |
| Output / 1M | $0.35 |
Compare
PaddleOCR-VL 1.6 is one of the catalog's OCR models. The table is each model's published price, size, and context.
| Model | What it does | Price | Parameters | Context |
|---|---|---|---|---|
| PaddleOCR-VL 1.6 | PaddleOCR-VL-1.6 for OCR, tables, formulas, and charts. | $0.15 / 1M input | 0.9B | 16K |
| PP-OCRv6 | PaddleOCR PP-OCRv6 medium text detection and recognition; scene OCR JSONL on image chat; text returns plain text; document_url fan-out via text. | $0.02 / 1M input | 20M | — |
| DeepSeek-OCR-2 | DeepSeek OCR 2 and markdown extraction. | $0.25 / 1M input | 3.4B | 32K |
| dots.mocr | Multilingual document layout parsing and markdown OCR. | $0.20 / 1M input | 3B | 32K |
Benchmarks
Published scores for PaddleOCR-VL 1.6, from the OmniDocBench leaderboard.
- OmniDocBench v1.696.34
- Formula CDM97.53
- Table TEDS94.76
| Reading | Example |
|---|---|
| Qualitative | Clear structure, grounded in the input |
| Quantitative | OmniDocBench v1.6: 96.34. |
| Cost and performance | Lower listed rate, mid-pack latency |
Performance
Latency is the end-to-end round trip. Throughput is completion tokens divided by that time. It is a request rate, not decode speed.
Latency
625 ms
Throughput
—
Requests
355
Supported inputs
- textIgnored
A text part beside the media is accepted and ignored. A request with no media part is a
400. - image_urlYes
One per request. A second
image_urlpart is a400. - video_urlNo
Not accepted.
- document_urlYes
One PDF, up to 128 pages, rasterized per page at
document_dpi(default 96). See Document Inputs.
- Default method
markdown. All five methods return a text payload; only the subject differs, sotable,formula, andchartread a page's tables, equations, and charts rather than its prose. - Context window 16,384 tokens.
- Aliases:
paddleocr-vl-1.6,PaddlePaddle/PaddleOCR-VL-1.6, and the 1.5-era idspaddlepaddle/paddleocr-vl,paddleocr-vl,PaddlePaddle/PaddleOCR-VL. - Weights:
PaddlePaddle/PaddleOCR-VL-1.6.
Methods
Methods this model serves. Payload shapes are in the docs.
| Method | Returns |
|---|---|
| markdown (default) | Reading-order page read as Markdown |
| ocr | Reading-order plain-text page read |
| table | The page's table alone, as OTSL structure tokens |
| formula | LaTeX for the page's equations (e.g. $$E = mc^2$$) |
| chart | Short description followed by the series as a Markdown table |
Note
A gateway build that predates the catalog fix lists only ocr, table, formula, and chart for this model and reports default_method as ocr, while still applying markdown when method is omitted. Send method explicitly whenever the page read matters.
Every method on this model uses the shared contract on Methods & Response Format: <document> / <page id= format= width= height=> in text mode, and the flat reply with top-level model, method and the input's details in JSON mode. The method decides the payload, not the shape.
Estimate cost
Estimate. A page is 2,500 input tokens and 1,000 output tokens.
$0.73
Quick start
Call this model on the OpenAI-compatible gateway. The model id is already filled in.
Swap image_url for a document_url PDF to read every page with the same method; markdown is used when method is omitted.
Response
Both tabs below show method: "table", the method the request above uses. Every method on this model follows the same shared contract, so swapping in markdown (or omitting method) changes the payload, not its shape.
A single image returns the payload alone, with no wrapper. The table payload is OTSL structure tokens, where <fcel> opens a cell, <lcel> continues one to the left, and <nl> ends a row:
| Metric | 2023 | 2024 |
| Revenue | 0.9M | 1.2M |
A PDF returns one <document> block per file and one <page> block per page:
Page 0
| UNITED STATES SECURITIES AND EXCHANGE COMMISSION |
…
