Download
Dokumentli® whitepaper and product sheet.
Two documents in one package. The whitepaper explains what a small special-purpose vision language model is, shows the benchmark results against five general-purpose frontier models and states plainly what the model is not built for. The product sheet gives you technical requirements, API and pricing on four pages.
89.1%
average accuracy (cANLS@0.8), the highest score among six tested models
5 benchmarks, 3,365 documents, as of 25.09.20261.9x
faster than the fastest reference model
1.14 documents per second1 GPU
from 24 GB VRAM is enough to run the full model, air-gapped if needed
no data centre, no cloud connection requiredGet both PDFs
Fill in the form and the downloads appear right away.
Contents
What is actually in the two documents.
Not a brochure. Numbers, architecture, requirements, pricing.
Whitepaper, 11 pages
What a small special-purpose VLM isBenchmark results, accuracyBenchmark results, speedWhy specialisation wins hereHonest limitationsFour deployment optionsTypical integration patternsCost logic against general-purpose APIs
Product sheet, 4 pages
Key figures at a glanceDokumentli Node, monthly pricingPage volume per packageHardware requirementsLinux, NVIDIA from 24 GB VRAM, CUDAOpenAI-compatible chat completions APIDeployment variants including air-gappedContact and next steps
Who this is built for: European companies and regulated industries with high, recurring document volume. From financial services and insurance to construction, trade and logistics.
Rather test it yourself
The documents do not replace your own test.
If you want to know how Dokumentli reads your own documents, upload them in the playground. No installation, no form.
