{"name":"table-formats","title":"Table formats","tables":[{"heading":"How each engine writes a table (official docs, model cards and code, read Oct 3 2026)","columns":["Engine","Table format","Merged cells","Can you change it?","Source"],"rows":[["Landing AI DPT-3","HTML inside the Markdown; every cell also as JSON with its row, column and span","Kept as rowspan and colspan; in Markdown mode they expand into empty cells","Yes: html (default) or markdown","Changelog: HTML tables by default since Jun 26 2026"],["Reducto","Markdown for small tables (30 cells or fewer, 4 merged or fewer), HTML for larger ones","Kept in HTML only","Yes: dynamic (default), html, md, json, csv or jsonbbox","Current docs, undated"],["Mistral OCR 4.1","Markdown, inline in the page","Not stated","Yes: inline Markdown (default), or markdown or html returned beside the page","Current docs, undated"],["HunyuanOCR-1.5 (Tencent)","HTML inside the Markdown, by its default document-parsing prompt","Not stated","Through the prompt","Card updated Aug 28 2026"],["OvisOCR2 (Alibaba)","HTML inside the Markdown","Not stated","No: one fixed prompt","Card updated Aug 28 2026"],["dots.ocr (rednote)","HTML","Not stated","No table option","Released Jul 30 2025"],["TeleOCR (China Telecom)","OTSL, turned into an HTML table by its own converter","Kept: OTSL's span tokens become rowspan and colspan","Per-element prompts","Card updated Sep 29 2026"],["qwen3.5-ocr / qwen-vl-ocr (Alibaba)","HTML in its table-parsing task; tables in its document-parsing task not stated","Asked for in the prompt; how they are written not stated","Yes: choose the task, or write your own prompt","Docs updated Sep 28 2026"],["Qianfan-OCR (Baidu)","HTML in its table-recognition task; its full-page Markdown parse shows no table","Kept, per its card","Through the prompt only","Released Mar 18 2026"],["PaddleOCR-VL-1.6 (Baidu)","HTML inside the Markdown: the model writes OTSL, and its pipeline turns it into an HTML table","Kept as rowspan and colspan","No table-format option found","Pipeline code, PaddleX 3.7.2"],["GLM-OCR (Zhipu)","Not stated","Not stated","A table-recognition prompt; no format option","Paper Mar 11 2026"],["DeepSeek-OCR 2","Not stated","Not stated","Two prompts only","Paper Jan 28 2026"],["Unlimited-OCR (Baidu)","Not stated","Not stated","Prompts only","Release note Jul 21 2026"],["MinerU2.5-Pro","Not stated for tables; MinerU 4.0 renders whole documents to Markdown, HTML, LaTeX, DOCX and more","Not stated","Not found","Current README, undated"]],"note":"Not stated means the official page doesn't say; nothing here is guessed. Mistral's table_format option returns tables beside the page and leaves a placeholder in its Markdown."},{"heading":"A full page's tables, by default","columns":["Format","Engines","Which"],"rows":[["HTML","4","Landing AI DPT-3, HunyuanOCR-1.5, OvisOCR2, dots.ocr"],["HTML for larger tables, Markdown for small ones","1","Reducto"],["OTSL from the model, turned into HTML by its own code","2","TeleOCR, PaddleOCR-VL-1.6"],["Markdown","1","Mistral OCR 4.1"],["HTML in a separate table task; a full page not stated","2","Qianfan-OCR, qwen3.5-ocr / qwen-vl-ocr"],["Not stated","4","GLM-OCR, DeepSeek-OCR 2, Unlimited-OCR, MinerU2.5-Pro"]]}],"log":[{"date":"261004","line":"New board. 7 of the 8 engines whose full-page table format is known return HTML tables; Mistral OCR 4.1 defaults to Markdown."}],"updatedAt":"2026-10-04T01:04:17.295Z"}