Demist-2

Training compute

4.6×10²⁰ FLOP

Parameters

95M

Published

Apr 17, 2025

Demist-2 is an AI model developed by Darktrace (United Kingdom), first published in April 2025. It works in the language domain, on tasks such as language modeling/generation, text classification, question answering and entity embedding.

Training it took an estimated 4.6×10²⁰ FLOP of compute (estimation method: operation counting,hardware). The model has 95,000,000 parameters. It was trained on roughly 351B datapoints. Training ran on 8 NVIDIA A100 for about 216 hours.

Access: Hosted access (no API). Its weights are not openly released. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Darktrace
Country of organization: United Kingdom
Domain: Language
Task: Language modeling/generation, Text classification, Question answering, Entity embedding
Training compute: 4.6×10²⁰ FLOP
Compute estimation method: Operation counting, Hardware
Parameters: 95,000,000
Dataset size: 351B
Training hardware: NVIDIA A100
Chips used: 8
Training time: 216 h
Training power draw: 6.3 kW
Model accessibility: Hosted access (no API)
Open weights: No
Epoch confidence: Confident