Model reference — Documents
Text classification sorts each piece of text into one of your own categories — a support ticket into a queue, an email into a risk band, a product review into a theme — and tagging attaches several labels to the same text where more than one applies.
There are two routes. A small purpose-built classifier is fitted to your categories using examples you already have; it is cheap to run, fast, and the most accurate option once a few hundred examples per category exist. A large language model classifies from a written description of each category with no examples at all, which is the practical answer when the categories are new or change often. Many deployments start on the second route and move to the first once labelled data has accumulated.
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 document of about 500 words assigned to one or more categories.
Vendor: Microsoft
What it does: a small classifier fitted to your categories from labelled examples. Among the most accurate models of its size on English text, and cheap enough to classify millions of documents a day.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | 1× 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 (documents/hour) | ≈ 20,000 | ≈ 90,000 | ≈ 260,000 |
Vendor: Meta
What it does: a well-proven classifier of the same family, widely used and easy to fit. A dependable baseline against which any newer model should be measured.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | 1× 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 (documents/hour) | ≈ 24,000 | ≈ 100,000 | ≈ 300,000 |
Vendor: Meta
What it does: the same approach trained across a hundred languages, so one fitted classifier handles text arriving in many languages without a separate model for each.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | 1× A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 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 (documents/hour) | ≈ 18,000 | ≈ 80,000 | ≈ 230,000 |
Vendor: Hugging Face
What it does: fits a working classifier from as few as eight examples per category, which makes it the fastest way to get a new category live. Accuracy then improves as more labelled examples arrive.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | 1× 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 (documents/hour) | ≈ 22,000 | ≈ 95,000 | ≈ 280,000 |
Vendor: Beijing Academy of Artificial Intelligence
What it does: turns each text into a numeric fingerprint that a simple classifier sits on top of. The advantage is that one fingerprinting pass serves classification, search and duplicate detection at the same time.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | 1× 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 (documents/hour) | ≈ 6,000 | ≈ 30,000 | ≈ 90,000 |
Vendor: Meta
What it does: classifies from a written description of each category, with no labelled examples needed, and explains its choice. The practical route for categories that are new, numerous or still being argued about.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | 1× 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 (documents/hour) | ≈ 1,400 | ≈ 4,000 | ≈ 14,000 |
Vendor: Mistral AI
What it does: the same instruction-driven approach with better handling of long documents and subtle category definitions, at a moderate cost per document.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 4090 | 1× L40S 48 GB | 1× 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 (documents/hour) | ≈ 900 | ≈ 2,600 | ≈ 8,000 |
The choice depends on how many categories you have, how stable they are and how much labelled text exists. Our consultants review your categories and your data, then recommend either a fitted classifier or a language model, and say what accuracy each will reach on your material.
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 Text classification / tagging. 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.