Introducing Mistral OCR 3 | Mistral AI

Highlights

• Breakthrough performance: 74% overall win rate over Mistral OCR 2 on forms, scanned documents, complex tables, and handwriting.

• State-of-the-art accuracy, outperforming each enterprise document processing solutions in addition to AI-native OCR solutions

• Now powers Document AI Playground in Mistral AI Studio, a straightforward drag-and-drop interface for parsing PDFs/images into clean text or structured JSON

• Major upgrade over Mistral OCR 2 in forms, handwritten content, low-quality scans, and tables

Overview

Mistral OCR 3 is designed to extract text and embedded images from a wide selection of documents with exceptional fidelity. It supports markdown output enriched with HTML-based table reconstruction, enabling downstream systems to know not only document content, but additionally structure. As a much smaller model than best solutions, it is on the market at an industry-leading price of $2 per 1,000 pages, with a 50% Batch-API discount, reducing the fee to $1 per 1,000 pages.

Developers can integrate the model (mistral-ocr-2512) via API, and users can leverage Document AI, a UI that parses documents into text or structured JSON immediately.

Complex TablesForm understandingLetter

Complex Tables4941 chars

RenderMarkdown

TABLE 21. Doctoral degrees awarded to men, by major field group: 1966-2012

Academic yr ending	All fields	Science and engineering fields								Non-S&E fields
		Total	Biological and agricultural sciences	Earth, atmospheric, and ocean sciences	Mathematics and computer sciences	Physical sciences	Psychology	Social sciences	Engineering
1966	15,863	10,646	2,386	392	722	2,535	894	1,424	2,293	5,217
1967	17,961	12,013	2,565	412	782	2,930	1,030	1,699	2,595	5,948
1968	20,005	13,328	3,028	432	924	3,064	1,131	1,906	2,843	6,677
1969	22,355	14,781	3,278	487	1,013	3,241	1,350	2,156	3,256	7,574
1970	25,527	16,404	3,627	493	1,148	3,666	1,446	2,604	3,420	9,123
1971	27,271	17,385	3,897	538	1,142	3,718	1,615	2,992	3,483	9,886
1972	27,754	17,191	3,802	561	1,185	3,404	1,670	3,088	3,481	10,563
1973	27,670	16,853	3,764	557	1,113	3,209	1,741	3,151	3,318	10,817
1974	26,594	16,043	3,571	547	1,096	2,902	1,797	3,016	3,114	10,551
1975	25,751	15,870	3,623	535	1,038	2,811	1,878	3,035	2,950	9,881
1976	25,262	15,375	3,559	531	890	2,617	1,937	3,061	2,780	9,887
1977	23,858	14,775	3,470	588	837	2,477	1,902	2,932	2,569	9,083
1978	22,553	14,199	3,449	524	828	2,364	1,928	2,736	2,370	8,354
1979	22,301	14,128	3,516	542	833	2,381	1,831	2,596	2,429	8,173
1980	21,612	13,814	3,599	530	846	2,199	1,787	2,464	2,389	7,798
1981	21,463	14,056	3,607	484	822	2,318	1,885	2,511	2,429	7,407
1982	21,016	13,924	3,594	512	824	2,337	1,720	2,415	2,522	7,092
1983	20,748	13,920	3,429	501	838	2,430	1,750	2,315	2,657	6,828
1984	20,636	13,954	3,565	472	841	2,446	1,625	2,244	2,761	6,682
1985	20,552	14,043	3,530	470	859	2,452	1,576	2,188	2,968	6,509
1986	20,592	14,268	3,378	462	959	2,585	1,527	2,207	3,150	6,324
1987	20,934	14,580	3,307	491	999	2,686	1,474	2,153	3,470	6,354
1988	21,677	15,267	3,477	541	1,087	2,759	1,392	2,111	3,900	6,410
1989	21,811	15,622	3,481	542	1,209	2,627	1,408	2,188	4,167	6,189
1990	22,960	16,498	3,680	581	1,329	2,840	1,368	2,221	4,479	6,462
1991	23,521	16,982	3,743	626	1,514	2,919	1,249	2,229	4,702	6,539
1992	24,235	17,423	3,816	585	1,588	2,961	1,327	2,285	4,861	6,812
1993	24,387	17,571	3,800	566	1,602	2,868	1,322	2,315	5,098	6,816
1994	25,061	18,167	3,940	622	1,641	3,104	1,273	2,435	5,152	6,894
1995	25,162	18,119	3,989	577	1,727	2,922	1,245	2,388	5,271	7,043
1996	25,293	18,461	4,101	565	1,656	2,961	1,163	2,523	5,492	6,832
1997	24,944	18,084	4,046	608	1,594	2,880	1,162	2,478	5,316	6,860
1998	24,630	17,810	4,075	564	1,638	2,865	1,205	2,352	5,111	6,820
1999	23,439	16,735	3,919	535	1,495	2,722	1,209	2,350	4,505	6,704
2000	23,166	16,518	3,943	495	1,507	2,546	1,203	2,365	4,459	6,648
2001	22,782	16,189	3,766	461	1,407	2,531	1,128	2,324	4,572	6,593
2002	21,812	15,392	3,830	477	1,291	2,335	1,065	2,217	4,177	6,420
2003	22,257	15,761	3,777	470	1,419	2,396	1,043	2,286	4,370	6,496
2004	22,965	16,417	3,830	448	1,520	2,470	1,081	2,313	4,755	6,548
2005	23,736	17,405	3,913	470	1,782	2,670	1,058	2,285	5,227	6,331
2006	25,023	18,375	3,998	490	2,074	2,831	936	2,317	5,729	6,648
2007	26,203	19,542	4,388	542	2,308	2,905	941	2,317	6,141	6,661
2008	26,272	19,857	4,489	548	2,353	2,957	999	2,343	6,168	6,415
2009	26,332	19,840	4,495	539	2,327	2,995	991	2,487	6,006	6,492
2010	25,527	19,570	4,355	496	2,435	2,924	1,033	2,523	5,804	5,957
2011	26,192	20,380	4,464	522	2,494	3,167	1,003	2,527	6,203	5,812
2012	27,390	21,233	4,578	496	2,683	3,214	1,047	2,688	6,527	6,157

S&E = science and engineering.
NOTE: See appendix B for specific fields which are included in each category.
SOURCE: National Science Foundation, National Center for Science and Engineering Statistics, Survey of Earned Doctorates.

Benchmarks

To lift the bar, we introduced tougher internal benchmarks based on real business use-case examples from customers. We then evaluated several models across the domains highlighted below, comparing their outputs to ground truth using fuzzy-match metric for accuracy.

Upgrades over previous generations of OCR models

Whereas most OCR solutions today focus on specific document types, Mistral OCR 3 is designed to excel at processing the overwhelming majority of document types in organizations and on a regular basis settings.

• Handwriting: Mistral OCR accurately interprets cursive, mixed-content annotations, and handwritten text layered over printed forms.

• Forms: Improved detection of boxes, labels, handwritten entries, and dense layouts. Works well on invoices, receipts, compliance forms, government documents, and such.

• Scanned & complex documents: Significantly more robust to compression artifacts, skew, distortion, low DPI, and background noise.

• Complex tables: Reconstructs table structures with headers, merged cells, multi-row blocks, and column hierarchies. Outputs HTML table tags with colspan/rowspan to totally preserve layout.

Mistral OCR 3 is a major upgrade across all languages and document form aspects in comparison with Mistral OCR 2. 

Recommend use cases and applications

Mistral OCR 3 is good for each high-volume enterprise pipelines and interactive document workflows. Developers can use it for:

• Extracting text and pictures into markdown for downstream agents and knowledge systems

• Automated parsing of forms, invoices, and operational documents

• End-to-end document understanding pipelines

• Digitization of handwritten or historical documents

• Another document → knowledge transformation applications. 

Our early customers are using Mistral OCR 3 to process invoices into structured fields, digitize company archives, extract clean text from technical and scientific reports, and improve enterprise search. 

Available today 

Access the model either through the API or via the brand new Document AI Playground interface, each in Mistral AI Studio. Mistral OCR 3 is fully backward compatible with Mistral OCR 2. For more details, head over to mistral.ai/docs.