Named entity recognition finds the things a document is about and labels them — people, companies, places, dates, amounts, contract numbers, drug names, part codes — and returns each one with its position in the text.
It is the step that turns prose into data. Once entities are labelled, documents can be indexed by the companies they mention, matched against a watchlist, or checked for the presence of personal information. Off-the-shelf models cover the common types well; the value in most projects lies in the types specific to your business, which either need examples to fit a model to, or a model that accepts a written list of what to look for.
Models in this group take text or whole documents as input: plain text, PDFs, scanned pages and office files. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is the number of sample inputs processed per hour on that hardware. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one A4 page of about 500 words with all entities labelled.
GLiNER
Vendor: Urchade Zaratiana and contributors
What it does: finds any entity type you name in plain words — "policy number", "vessel name", "active ingredient" — with no training data at all. The fastest route to covering entity types no off-the-shelf model knows.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
6 GB
24 GB
80 GB
vCPUs
4
8
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (pages/hour)
≈ 3,000
≈ 12,000
≈ 34,000
spaCy transformer pipelines
Vendor: Explosion
What it does: labels the standard types — people, organisations, places, dates, money — in a production-grade pipeline with sentence splitting and part-of-speech tagging included. Well documented and dependable for bulk processing.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
6 GB
24 GB
80 GB
vCPUs
4
8
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (pages/hour)
≈ 12,000
≈ 45,000
≈ 130,000
DeBERTa v3
Vendor: Microsoft
What it does: fitted to your own entity types from labelled examples. The most accurate option once a few hundred examples of each type exist, and cheap enough to run over an entire archive.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
6 GB
24 GB
80 GB
vCPUs
4
8
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (pages/hour)
≈ 8,000
≈ 30,000
≈ 90,000
XLM-RoBERTa
Vendor: Meta
What it does: labels entities across a hundred languages with one model, so names and amounts are found in documents whatever language they arrive in.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
8 GB
24 GB
80 GB
vCPUs
6
12
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (pages/hour)
≈ 6,000
≈ 24,000
≈ 70,000
Llama 3.1 8B
Vendor: Meta
What it does: extracts entities from a written description and copes with unusual phrasing and context — telling a person’s name from a street named after them. Slower per page, so it is usually reserved for documents a small model got wrong.
Requirement
Minimum
Medium
High
GPU type
RTX 3090
RTX 4090
H100 80 GB
VRAM
16 GB
24 GB
80 GB
vCPUs
8
12
24
RAM
32 GB
48 GB
128 GB
Server
1× RTX 3090 24 GB
1× RTX 4090 24 GB
1× H100 SXM 80 GB
Rate (pages/hour)
≈ 900
≈ 2,800
≈ 9,000
Mistral Small 3
Vendor: Mistral AI
What it does: the same instruction-driven extraction with better handling of long documents, and able to return entities already grouped — every mention of the same company resolved to one record.
Requirement
Minimum
Medium
High
GPU type
RTX 4090
L40S 48 GB
H100 80 GB
VRAM
24 GB
48 GB
80 GB
vCPUs
12
16
32
RAM
48 GB
64 GB
128 GB
Server
1× RTX 4090 24 GB
1× L40S 48 GB
1× H100 SXM 80 GB
Rate (pages/hour)
≈ 700
≈ 2,000
≈ 6,000
Choosing between them
The question is whether your entity types are standard, and how much labelled text you have. Our consultants review your documents and the entities you need, then recommend either an instruction-driven model that needs no training data or a fitted model that will be more accurate on your own types.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Named entity recognition. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.