LightOn
LightOn is a French AI company founded in 2016 that provides document parsing, extraction, and retrieval APIs for enterprise applications.
Profile
LightOn provides a multimodal retrieval API that parses, extracts, and searches documents, built on open-source models and designed for AI agent workflows.
LightOn is a French AI company founded in 2016 that provides document parsing, extraction, and retrieval APIs for enterprise applications. The company went public on Euronext Growth Paris in November 2024 (ticker: ALTAI-FR) and reported a 15% rise in first-half 2025 revenue compared to the prior year, though it posted a net loss of €3.7 million after investing roughly €6 million in infrastructure and hiring. In March 2026, LightOn reported its FY 2025 results, and in April 2026 it strengthened its cash position with a bond financing of up to €4 million in nominal value.
The company's product portfolio centers on three API endpoints: /parse (OCR and document-to-Markdown conversion, with its LightOnOCR-2 model achieving 83.2 on the OLMOCR-BENCH benchmark at €0.002 per page), /extract (schema-defined field extraction from documents at €0.004 per page), and /search (hybrid retrieval combining dense, sparse, and late-interaction signals at €0.006 per query with sub-200ms P50 latency). LightOn emphasizes that its retrieval layer is built for AI agents, not just chatbots, and supports MCP-native integration, workspace-level access control, and LLM-agnostic deployment. The company maintains a significant open-source presence: its models (including LateOn, NextPlaid, PyLate, DenseOn, and LightOnOCR-2) have accumulated over 50 million HuggingFace downloads, 916,000 monthly PYPI installs, and 2,345 GitHub stars.
In early 2026, LightOn launched a sovereign SaaS offering aimed at simplifying enterprise integration of document AI, and in May 2026 it signed an agreement with Infocom'94 to accelerate sovereign generative AI deployment for local authorities in France. The company's investors include historical backers and founders, with public shareholders holding a portion of the capital. LightOn faces the challenge of translating its open-source community traction into sustained commercial revenue while competing with larger players in the document AI and retrieval-augmented generation space.
Products by LightOn
- D DenseOn DenseOn is an open-source dense retrieval model developed by LightOn, designed for high-accuracy semantic search and document retrieval. Be the first to review →
- L LightOnOCR-2 LightOnOCR-2 is an end-to-end 1B-parameter vision-language model designed for high-performance optical character recognition (OCR) tasks, targeting enterprises and Be the first to review →
Who buys this
- Enterprises needing document processing (OCR, field extraction) for invoices, contracts, and forms
- Developers building retrieval-augmented generation (RAG) systems for AI agents
- Government and local authorities requiring sovereign AI solutions for document management
- Organizations with multi-agent systems needing scoped corpora and chunk-level access control
Strengths and what to watch
Strengths
- Strong open-source ecosystem: 50M+ HuggingFace downloads, 916K monthly PYPI installs, and 2,345 GitHub stars for models like LateOn and NextPlaid
- Publicly traded on Euronext Growth Paris with regulated financial disclosures, providing transparency on revenue (H1 2025 up 15%) and cash position (€4M bond financing in April 2026)
- Sovereign SaaS offering and partnership with Infocom'94 for local authorities in France, differentiating from US-based competitors on data residency and compliance
Watch for
- Net loss of €3.7 million in H1 2025 despite revenue growth, with ongoing investment in infrastructure and hiring that may delay profitability
- Dependence on open-source community engagement for brand and adoption; commercial revenue from API endpoints may not yet match community traction
- Competition from larger document AI and RAG platforms (e.g., Google, Microsoft, and well-funded startups) that offer similar parsing and retrieval capabilities
Recent moves
- 2mo ago LightOn Signs Agreement with Infocom'94 to Accelerate the Deployment of Sovereign Generative AI for Local Authorities
- 3mo ago LightOn strengthens its cash position with a bond financing of up to €4 million in nominal value
- 4mo ago FY 2025 Results
- 6mo ago LightOn launches a sovereign SaaS offering to simplify enterprise integration of document AI
- 10mo ago France's LightOn lifts H1 revenue by 15%
Key Information
- Industry
- Commercial AI Research
- Founded
- 2016
Frequently Asked Questions
What does LightOn do?
LightOn provides a multimodal retrieval API for parsing, extracting, and searching documents. Built on open-source models, it is designed for AI agent workflows, helping enterprises process invoices, contracts, and forms with OCR and field extraction.
What are LightOn's main API endpoints and their pricing?
LightOn offers three API endpoints: /parse for OCR at €0.002 per page, /extract for schema-defined field extraction at €0.004 per page, and /search for hybrid retrieval at €0.006 per query with sub-200ms latency.
How does LightOn's OCR model perform on benchmarks?
LightOn's LightOnOCR-2 model achieves a score of 83.2 on the OLMOCR-BENCH benchmark. It is used in the /parse endpoint for document-to-Markdown conversion and OCR at a cost of €0.002 per page.
Is LightOn a publicly traded company?
Yes, LightOn went public on Euronext Growth Paris in November 2024 under the ticker ALTAI-FR. It reported a 15% rise in first-half 2025 revenue but posted a net loss of €3.7 million due to infrastructure investments.
What open-source models has LightOn released?
LightOn has released models like LateOn, NextPlaid, PyLate, DenseOn, and LightOnOCR-2. These have accumulated over 50 million HuggingFace downloads, 916,000 monthly PYPI installs, and 2,345 GitHub stars.
What is LightOn's sovereign AI offering for local authorities?
In early 2026, LightOn launched a sovereign SaaS offering for enterprise document AI. In May 2026, it signed an agreement with Infocom'94 to accelerate sovereign generative AI deployment for local authorities in France, focusing on data residency.
Sources
- www.lighton.ai — Product details: three API endpoints (parse, extract, search), pricing, open-source model stats (50M HF downloads, 916K PYPI installs, 2,345 GitHub stars), LightOnOCR-2 benchmark score (83.2 on OLMOCR-BENCH)
- www.lighton.ai — IPO date (Nov 26, 2024), ticker (ALTAI-FR), bond financing (€4M, April 2026), sovereign SaaS launch (Feb 2026), Infocom'94 agreement (May 2026), FY 2025 results (March 2026)
- www.reuters.com — H1 2025 revenue up 15%, net loss of €3.7 million, €6M investment in infrastructure and hiring
- techcrunch.com — General tech industry context; no specific LightOn article found but used to confirm no recent TechCrunch coverage of LightOn