Demist-2

Closed weights Darktrace 95M parameters April 2025

No estimate

No hardware requirements for this model

The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
Darktrace
Organisation type
Industry
Country
United Kingdom of Great Britain and Northern Ireland
Published
17 April 2025
Authors
Miguel Neves Fonseca, Philip Sellars, Tim Bazalgette

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Language
Task
Language modeling/generation, Text classification, Question answering, Entity embedding

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
95M

95M

Training data
351,000,000,000 tokens

C4: We extracted a 305 billion token subset URIs: We extracted the available URIs from several Common Crawl web scrapes. This allowed us to create a large corpus of 10 billion tokens. Linked Hostnames: Using the Common Crawl dataset, we extracted the observed links between web pages. <..> We further processed this data into rows of hostnames that shared linking properties which totaled 11 billion tokens. HTTP Data: 25 billion tokens. 305 + 10 + 11 + 25 = 351 b tokens "1.2 million optimization …

Epochs
1.8

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute
4.6 × 10²⁰ FLOP

6 FLOP / parameter / token * 95 * 10^6 parameters * 351 * 10^9 tokens [see dataset size notes] *1.8 epochs = 3.60126e+20 FLOP 312000000000000 FLOP / GPU / sec [A100 reported] * 8 GPUs * 216 hours [9 days reported] * 3600 sec / hour * 0.3 [assumed utilization] = 5.8226688e+20 FLOP sqrt(3.60126e+20 * 5.8226688e+20) = 4.579186e+20 FLOP

How it was established
Operation counting,Hardware

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Training hardware
NVIDIA A100
Chips used
8
Wall-clock time
216 hours (9 days)

"DEMIST-2 was trained on 8 A100 GPUs over a nine-day period using the AWS Sagemaker platform." 9 days = 216 hours

Power draw
6.3 kW
Cloud vendor
AWS,Amazon Web Services

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights
Closed — provider access only
Model access
Hosted access (no API)
Training code
Unreleased

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Record confidence
Confident

Sources

Where this record came from and when it was last checked.

Reference
DEMIST-2: Darktrace Embedding Model for Investigation of Security Threats
Last updated
28 November 2025

What the numbers mean

Where it came from

Demist-2 was published by Darktrace, in United Kingdom of Great Britain and Northern Ireland, in April 2025. The organisation is categorised as industry.

It works in Language, and is recorded as doing language modeling/generation, Text classification, Question answering, Entity embedding.

Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.

Training and provenance

Training it took roughly 4.6 × 10²⁰ FLOP of computation, on NVIDIA A100 — a measure of what producing the model cost, not of how fast it answers.

Around 351,000,000,000 tokens went into training it.

Answers

Demist-2 — common questions

01

Who created Demist-2?

Demist-2 was published by Darktrace, based in United Kingdom of Great Britain and Northern Ireland, categorised as industry.

02

When was Demist-2 released?

Demist-2 was published in April 2025.

03

What is Demist-2 used for?

Demist-2 works in Language, and is recorded as handling language modeling/generation, Text classification, Question answering, Entity embedding. These are the areas it was designed around; they describe intent rather than a hard boundary.

04

How much compute was used to train Demist-2?

Around 4.6 × 10²⁰ FLOP, on NVIDIA A100. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

05

What GPU do I need to run Demist-2?

None. Demist-2 is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.

06

Is Demist-2 open source?

No. Demist-2 has not had its weights published, so it exists only as a service controlled by its owner.

07

How many parameters does Demist-2 have?

Demist-2 has 95M parameters. 95M. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

Source

Original publication

Record last updated 28 November 2025

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.