LightOn (Paradigm)

Sovereign on-prem GenAI platform for regulated European enterprises

LLMs & Chat Enterprise Has API
Researched · Published · Reviewed
RECATOOLS Score
6.1 / 10
Capability
6
Value for money
6
Ease of use
6
ASEAN readiness
5
API quality
7
Founded
HQ
Users
Launched
Developer

Overview

LightOn is a Paris enterprise-AI firm, the first GenAI company listed on Euronext Growth Paris (Nov 2024). Paradigm, its flagship platform, deploys on-premise or hybrid, bundling RAG, agents and curated open-weight models for banks, ministries and defense clients that need strict data residency.

Advertisement

What you can produce with LightOn (Paradigm)

  • On-premise or hybrid GenAI deployment for data-sovereignty needs
  • Paradigm: RAG + agents + multi-model access behind one API
  • Curated access to open-weight models (Mistral, Llama) plus in-house fine-tunes
  • Open-source models: PAGnol, ModernBERT variants, LightOnOCR-1B
  • PyLate retrieval library for late-interaction search
  • Public financial disclosure via Euronext Growth Paris listing
Advertisement

ASEAN Perspective

LightOn (Paradigm) in Southeast Asia

ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).

RECATOOLS Verdict

LightOn's pitch is straightforward: keep the model and the data on your own infrastructure. Paradigm bundles RAG, agents and access to curated open and in-house models behind one API, deployable on-prem or hybrid — the opposite of the API-only default most LLM vendors ship. Being the first GenAI company on Euronext Growth Paris (November 2024) forces a level of financial transparency you don't get from most enterprise-AI vendors: FY2025 revenue was up 54% to €1.7M, with Paradigm license sales roughly tripling.

Customers skew French public-sector and regulated-industry (Île-de-France, Safran, Groupama, France's tax authority), which tells you the target market — this is a sovereignty play, not a self-serve tool. There's no public pricing, no free trial and, so far, no meaningful independent review trail on G2 or Capterra. For ASEAN teams it's more useful as a blueprint for sovereign AI procurement than something to sign up for directly.

Independent AI-assisted assessment by RECATOOLS.

What people say

LightOn doesn't show up on the usual review sites — no G2 page, no Capterra listing, nothing meaningful on Trustpilot. That's typical for a sovereignty-focused enterprise platform sold through direct sales and RFPs rather than self-serve signup, but it does mean there's no crowd-sourced user sentiment to draw on here. What's public instead is financial and technical.

Founded in 2016, LightOn listed on Euronext Growth Paris in November 2024 — the first generative-AI company to do so — raising €11.9 million at a €62 million market cap, with Portzamparc/BNP Paribas as lead manager. That listing status forces disclosure most private AI vendors never publish: FY2025 revenue reached €1.7 million, up 54% year-on-year, and annual recurring revenue hit €1.9 million by December 2025, with Paradigm license sales nearly tripling. Named customers include Région Île-de-France, aerospace group Safran, insurer Groupama, France's Direction Générale des Finances Publiques (the tax authority), and the CNES space agency — a client list that reads as French public-sector and regulated-industry, not global self-serve.

The more telling signal for developers is LightOn's open-source output. Its GitHub org carries around 50 repositories and regularly ships models to Hugging Face: PAGnol (the first large-scale French generative model, 1.5B parameters), a family of ModernBERT variants tuned for enterprise retrieval, LightOnOCR-1B (a compact vision-language OCR model the company says beats larger general-purpose VLMs on accuracy per compute), and PyLate, a training and retrieval library for late-interaction search. That output gets genuine pickup in the French and European NLP research community, even if it doesn't generate G2-style star ratings.

The commercial pitch — RAG, agents and model access wrapped in an on-premise or hybrid deployment — competes less with OpenAI or Anthropic's API model and more with other sovereign-AI plays aimed at defense, banking and government procurement, where "the model never leaves our infrastructure" is the actual selling point, not raw benchmark scores. Pricing is entirely quote-based; nothing is published, and there's no self-serve trial to test it against.

Summary of public user & expert reviews, compiled by RECATOOLS.

About this listing

Researched on
Published on
Last reviewed

This entry was compiled from publicly available data including LightOn (Paradigm)'s official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with LightOn (Paradigm) unless explicitly stated.

Data accuracy

Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.

For the latest details, please refer to LightOn (Paradigm) directly →

Spotted something out of date? Suggest an update →

Advertisement