Hugo 32B · French law
Beats a frontier model. Half the size.
+10 %Accuracy
−50 %Parameters
−75 %Compute
14 daysTraining
Work
One model per trade.
Insurance
Assur
Opening early November 2026
Insurance law, explained to the person who holds the policy.
Law
Hugo 32B
FlexAI case study
French law, against a frontier model.
Marketing
Simulated audiences.
In development
Your audiences' answers and objections, simulated.
Healthcare
Data on site.
POC under way
A specialised model for a healthcare organisation in the Netherlands.
What they say
“It answers the way our own risk team answers.”
Head of Model RiskRetail bank, France
01 / 02
Method
Four steps, each agreed first.
We adapt the model to the sector first, without touching your data.
Our sideThe model learns the sector from public and synthetic material. Your data is never touched.
Your side, under licenceThe model moves into your environment. Your data stays there.
- IAuditScope the task.
- IIPilotTest against your baseline.
- IIIProductionConnect to your systems.
- IVDeploymentHost and monitor.
Ask it something
Illustrative.What does this exposure cost us in capital?
€4.2M.
Standardised approach: 100% risk weight on €52M drawn. Under IRB, at the counterparty's PD of 1.8%, €3.1M.
SourceArt. 194 CRR · collateral eligibility
“…the credit institution shall have the right to liquidate or retain, in a timely manner, the assets from which the protection derives…”
