Reason-ModernColBERT

Training compute

3.7×10²¹ FLOP

Parameters

150M

Published

May 22, 2025

Reason-ModernColBERT is an AI model developed by LightOn (France), first published in May 2025. It works in the language domain, on tasks such as language modeling/generation, question answering, quantitative reasoning and retrieval-augmented generation.

Training it took an estimated 3.7×10²¹ FLOP of compute (estimation method: operation counting). The model has 150,000,000 parameters.

Access: Open weights (non-commercial). Its weights are openly available. It is built on top of GTE-ModernColBERT-v1. Epoch AI rates the confidence of this record as likely.

Full record

Organization: LightOn
Country of organization: France
Domain: Language
Task: Language modeling/generation, Question answering, Quantitative reasoning, Retrieval-augmented generation
Training compute: 3.7×10²¹ FLOP
Compute estimation method: Operation counting
Parameters: 150,000,000
Training time: 2 h
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Base model: GTE-ModernColBERT-v1
Epoch confidence: Likely