Datalab
Datalab is a research lab and document-intelligence company that trains specialized models to turn difficult PDFs, scans, spreadsheets, slides, and images into accurate, machine-readable data. Its stated mission is to make knowledge accessible without trading off quality, transparency, or security.
At a Glance
- Frontier AI laboratories
- AI and software companies
- Fortune 500 enterprises
- Financial and legal teams
- +3 more
AI Tools by Datalab
(1)Chandra
Open OCR Model for Documents
Discussions
No discussions yet
Be the first to start a discussion about Datalab
Latest News
API submits and result checks became about twice as fast; Python SDK 0.9.1 released.
Datalab published a review of 15 document-parsing benchmarks and their reliability.
Surya 2.5 released with 1.7x the throughput of Surya 2 and the best open OCR score under 1B parameters on OmniParseBench.
Chandra 2.6 launched in every Datalab API mode, with higher accuracy and 2.5x faster median model processing than Chandra OCR 2.4.
Products & Services
Hosted document-intelligence API and Python SDK for converting, extracting, segmenting, evaluating, filling, and editing documents, with citations, confidence scores, and verifiable page/position references.
Document conversion processor that turns PDFs, Office files, scans, spreadsheets, and slides into Markdown, HTML, JSON, or chunks while preserving tables, equations, figures, layout, and reading order.
Schema-driven structured extraction processor that returns typed fields, source citations, confidence scores, reasoning/verification metadata, and multiple speed/accuracy modes.
Content-based packet segmentation that identifies document boundaries, labels document types, returns page ranges and confidence, and can feed downstream workflows.
Market Position
Datalab positions itself as a high-precision, model-building document-intelligence platform rather than a thin wrapper around a general LLM. Its specialized smaller models emphasize layout fidelity, multilingual OCR, traceability, speed, and lower cost; it competes with ABBYY, Azure Document Intelligence, AWS Textract, Google Document AI, Reducto, LlamaParse, Mistral OCR, and other OCR/document-parsing APIs while differentiating through open-source models, verifiable citations, and hosted or self-hosted deployment.
Leadership
Founders
Vik Paruchuri
Founder and CEO of Datalab; previously founded Dataquest, an online technical-skills education company, and was an early team member at edX (MIT). He created or led the open-source Marker and Surya projects and is a former competitive Kaggle participant.
Sandy Kwon
Datalab co-founder and former COO. Previously VP of Operations & GTM and a founding-team member at Sym, with earlier business operations and strategy roles at Nutanix and business/revenue operations experience at Algolia.
Executive Team
Vik Paruchuri
Founder and Chief Executive Officer
Founder of Dataquest and creator of Marker and Surya; previously an early team member at edX (MIT) and a competitive Kaggle participant.
Founding Story
Datalab was founded in Brooklyn in June 2024 after Vik Paruchuri's open-source work on Marker and Surya demonstrated demand for specialized document models. The founders set out to train smaller, domain-specific foundation models that could combine the accuracy and flexibility of modern AI with the speed, cost efficiency, traceability, and lower hallucination risk of traditional OCR; the company also wanted to make high-quality knowledge trapped in complex documents accessible.
Business Model
Revenue Model
Usage-based hosted API and SDK pricing charged per processor/page, with monthly subscriptions for teams, custom enterprise API or self-hosted/on-premise licensing, and paid commercial licensing for model weights beyond the free research/personal/startup terms.
Pricing Tiers
$20/$10 monthly included usage shown on the plan comparison, 25 requests/minute and 25 concurrency; hosted API/SDK, community support.
$400 included usage, 400 requests/minute and 400 concurrency, email and Slack support, BAA/DPA, and four custom-processor creations per month.
Custom processing solutions and terms, volume discounts, dedicated support, SSO, VPC or fully air-gapped deployment, and custom MSA/SLA.
Full-layout OCR returning text, tables, headings, lists, reading order, and layout boxes; no tiers or minimums.
Current Chandra API price stated in the Chandra 2.6 announcement.
Current accurate-mode price stated in the Chandra 2.6 announcement.
Target Markets
- Frontier AI laboratories
- AI and software companies
- Fortune 500 enterprises
- Financial and legal teams
- Universities and research organizations
- Government and regulated industries
- RAG indexing and embedding pipelines
- AI-agent document understanding and research
- Model-training data preparation
- Enterprise document processing and knowledge extraction
- Financial-statement and filing extraction
- Audit, compliance, and verifiable document workflows
- Anthropic
- ElevenLabs
- Legora
- Siemens Healthineers