KDL Frontier Parser Nano
KDLAI/KDL-Frontier-Parser-nano
published Jun 2026 · updated Jun 2026
KDL Frontier Parser Nano is a vision-language model that parses document images into structured text, tables, and layouts using a multi-step pipeline.
specs
| Task | Document Parsing |
| Architecture | Vision-Language Model (VLM) |
| Parameters | 1.2B |
| Context Length | 8192 tokens |
| Max Images Per Request | 1 |
best for
- ·Extracting structured table data from scanned invoices
- ·Converting document layouts into machine-readable text and metadata
- ·Parsing charts and figures from financial reports
- ·Automated data entry from forms and receipts
FAQ
It is designed for end-to-end document parsing: layout detection, text recognition, table and formula recognition, and chart analysis from page images.
It has 1.2B parameters, making it a lightweight nano tier. A larger ultra version is available through KoreaDeep.
It is an open-weight model, but no specific license is mentioned in the model card. Contact KoreaDeep for usage terms.
Use the gigarouter OpenAI-compatible endpoint with your API key. Send an image (URL or base64) and the appropriate task prompt (e.g., "\nTable Recognition:"). Set temperature=0 and skip_special_tokens=False.
It accepts page images (one per request). Supported tasks include layout detection, text, table, formula, and figure analysis, each with a fixed prompt.
We're benchmarking and onboarding KDL Frontier Parser Nano as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.