Which AI model to extract data from documents?
Invoices, resumes, tickets, forms: turning text into structured data (JSON).
GLM-5.3 Flash
- Index
- 151.9
- Price / M tokens
- €0.13 / €0.45
- Cost / 1 M
- ≈ €335
The cheapest model on the Pareto line: no model is both more capable and cheaper.
See in the ranking →Qwen3.8 27B
- Index
- 149.4
- Price / M tokens
- Free (local)
Open weights under the Apache 2.0 license, 27 billion parameters: it runs on a single server with a 24 GB GPU, and nothing leaves your premises.
See in the ranking →Claude Sonnet 5.5
- Index
- 165.0
- Price / M tokens
- €1.79 / €8.92
- Cost / 1 M
- ≈ €5,355
On the Pareto line, with one of the best capability scores: fewer errors to fix on complex or poorly scanned documents.
See in the ranking →Why these choices
For extraction, what matters: following a JSON format, inventing nothing, and the price per document at high volume. No benchmark in the index measures extraction directly: the capability index is a general guide, and price makes the difference at scale.
Take action
JSON schema: extract a U.S. invoice
JSON Schema to extract a U.S. invoice with an LLM: vendor, invoice number, dates, line items, subtotal, sales tax and total in USD.
JSON schema: extract a resume
JSON Schema to extract a resume with an LLM: experience, skills, education and certifications. Leaves out age, photo and other protected data.
Guardrail: mask U.S. personal data (SSN, phone, card)
Guardrail that masks U.S. personal data in AI assistant replies: SSNs, +1 phone numbers, personal emails and card numbers. Rules, 12 tests, code.
Private chat on your documents: Ollama, Qdrant, Open WebUI
docker-compose stack for a private chat on your documents: Ollama (model), Qdrant (vector database) and Open WebUI (interface).
Data: Epoch AI index (CC BY 4.0), developers' official prices converted at the ECB rate, updated 2026-10-12 · AI model ranking: intelligence, benchmarks and price