공시 • Jun 16
Lighton Demonstrates Flexibility of LightOnOCR-2 by Adapting It to Arabic Through Fine-Tuning LightOn demonstrated the flexibility of LightOnOCR-2, its document understanding model, by adapting it to Arabic through fine-tuning. This extension is based on an internal synthetic data generation pipeline designed to cover languages that remain underrepresented in the OCR tools currently available on the market. This demonstration is based on a dataset comprising 12,000 synthetic pages and their reference transcriptions, produced using a modified version of LightOn’s synthetic document generator. The corpus covers a wide range of document scenarios, including scanning artifacts, font variations, resolution levels, and document types. The output format remains the one used to train the bbox variant of LightOnOCR-2, with bounding box detection that associates text with its spatial location. Applying OCR to Arabic presents specific challenges. The script is written from right to left, characters are connected in cursive form, and open datasets, like specialized models, remain less widely available than for Latin-based languages. For organizations processing archives, administrative, legal, or heritage documents in Arabic, these limitations can slow the automation of document workflows. This demonstration is part of a broader effort to extend the model to a variety of domains, as reflected by its more than 3 million downloads and the fine-tunings already carried out by the community. It notably addresses needs encountered in the Middle East, where LightOn is already working with public- and private-sector organizations. This development is consistent with LightOn’s positioning: providing enterprise-grade generative AI building blocks that are open, controllable, and tailored to sensitive environments. LightOn is making the guides needed to reproduce this fine-tuning available on its Hugging Face space, with the aim of making this approach accessible to as many users as possible and adaptable to other document contexts. LightOnOCR-2 is released as open source under the Apache 2.0 license. It plays a central role in the production document ingestion process within LightOn Console, LightOn’s self-service offering. The open model and our production engine are therefore built on the same technological foundation. The base model achieves a score of 83.2% on OlmOCR-Bench. 공시 • Apr 23
LightOn SA, Annual General Meeting, May 28, 2026 LightOn SA, Annual General Meeting, May 28, 2026. Location: 19 place vendome, paris France Reported Earnings • Apr 06
Full year 2025 earnings released Full year 2025 results: Revenue: €5.32m (up 103% from FY 2024). Net loss: €7.39m (loss widened 56% from FY 2024). Revenue is forecast to grow 23% p.a. on average during the next 2 years, compared to a 4.7% growth forecast for the Software industry in France. New Risk • Feb 27
New major risk - Revenue and earnings growth Revenue has declined by 31% over the past year. This is considered a major risk. Ultimately, shareholders want to see a good return on their investment and that generally comes from sharing in the company's profits. If revenues are declining, then it is difficult for the company to prevent its earnings from declining as well. A trend of falling revenue can be very difficult to turn around. If the company is well already established it may also be a sign the company has matured and is in decline. In addition, if the company pays dividends it will also likely need to reduce or cut them, striking a dual blow to total shareholder returns. Currently, the following risks have been identified for the company: Major Risks Less than 1 year of cash runway based on free cash flow trend (-€4.5m free cash flow). Share price has been highly volatile over the past 3 months (17% average weekly change). Revenue has declined by 31% over the past year. Minor Risks Revenue is less than US$5m (€3.7m revenue, or US$4.4m). Market cap is less than US$100m (€35.3m market cap, or US$41.7m). 공시 • Jan 20
LightOn Opens a New Field for AI with LightOnOCR-2 LightOn integrates into Paradigm a technology that beats the world state of the art: LightOnOCR-2, capable of uncovering and structuring the information buried in all enterprise documents, including the most complex. This is the first of three phases "Bleu, Blanc, Rouge" that widen the gap between Paradigm and the rest of the market. Until now, the most critical enterprise documents such as sensitive contracts, technical files, regulatory archives have remained largely inaccessible to AI. Too complex, too sensitive, too voluminous. LightOn now makes it possible to exploit them, wherever they already reside. With one billion parameters, LightOnOCR-2 outperforms every competing model on the OlmOCR benchmark, including those nine times its size. World's best model for document intelligence, Light onOCR-2 is compact enough to be deployed on-premise, where the documents are. Its end-to-end architecture replaces traditional OCR pipelines and enables scaling. The next phases will focus on large-scale document search (Blanc), then comprehension models (Rouge). Once combined, these technologies will enable organizations to deploy documentary AI, integrated into their systems and fully sovereign, to make decisions grounded in all of their information. 
