Blog
Local models, document AI, and what they actually cost
Guides for European teams comparing on-prem models with OpenAI and Anthropic — cost, GDPR, and the workflows that should stay local.
- 8 September 20267 minLocal vs cloud
Local LLM vs OpenAI: the real cost crossover for document processing
A September 2026 cost model for invoice-style pages: when a local GPU undercuts GPT-4.1 and Claude Sonnet — and when the cloud API is still cheaper.
Read the article - 3 September 202610 minDocument processing
Clearing a 40,000-invoice AP backlog with local models: a worked example
A worked example of clearing an AI invoice backlog: 40,000 supplier invoices on local models — triage, schema, a 200-invoice pilot, validation, ERP delivery.
Read the article - 1 September 202611 minModel landscape
Sizing a GPU for document processing: memory class, card count and cost at 25,000, 100,000 and 500,000 pages a month
How to size a GPU for document processing: 24, 48 and 80 GB classes, a pages-per-hour formula, and buy-vs-rent costs in euros at 25k to 500k pages a month.
Read the article - 27 August 202612 minConsulting workflows
The hybrid AI stack: local models for 95% of your documents, frontier APIs for the exceptions
How a hybrid AI stack pairs local models with frontier APIs: the 95/5 split, €660 vs €1,100–1,800 a month at 100,000 pages, and rules for the 5% that leaves.
Read the article - 25 August 202611 minComparisons
Azure Document Intelligence and Textract alternatives for Europe: four options and when to switch
Azure Document Intelligence alternatives for Europe: EU-hosted APIs, open-weight models, managed pipelines and SaaS — and the page volumes where each one wins.
Read the article - 20 August 202610 minConsulting workflows
Which workflows should run on local models? A four-axis scoring method for on-prem vs cloud AI
A scoring method for which workflows should run on a local LLM and when to use on-prem vs cloud AI: volume, sensitivity, stability and availability, 0–3 each.
Read the article - 18 August 202610 minDocument processing
Invoice extraction at 100,000 pages a month: fixed cost per page versus per-token pricing
Invoice extraction cost per page at 100,000 pages a month: GPT-4.1, Claude Sonnet 5, a managed fixed-price pipeline and a DIY GPU, and what each leaves out.
Read the article - 13 August 202611 minGDPR & EU AI Act
EU AI Act obligations for document processing: mostly minimal risk, with one Annex III trap
EU AI Act document processing rules, plainly: extraction is mostly not high-risk, what Art. 4, 5 and 26 still ask of deployers, and the Annex III trap.
Read the article - 11 August 20269 minConsulting workflows
Stop paying OpenAI per invoice: a workflow-first method to reduce API costs
How to reduce OpenAI API costs without a model swap: inventory recurring workflows, measure tokens per unit, and route each to a local, mini or frontier model.
Read the article - 6 August 202613 minModel landscape
Best local OCR and document models in 2026: seven compared, matched to the pages you actually process
Best local OCR models 2026: Tesseract, PaddleOCR-VL, DeepSeek-OCR, Qwen3-VL, Gemma 4, Llama 4 Scout and Mistral OCR compared by task, license and GPU fit.
Read the article - 4 August 202612 minGDPR & EU AI Act
GDPR-compliant document AI: the five obligations, and the two a local model removes
What GDPR compliant document AI requires — lawful basis, a minimal schema, Art. 28 contracts, transfers, a DPIA — and which of those a local model removes.
Read the article - 30 July 202610 minDocument processing
AI document processing with local models: a buyer’s guide for European teams
How European buyers choose AI document processing with local models: four deployment options compared, honest accuracy tests, cost at three volumes, GDPR questions.
Read the article - 28 July 20269 minLocal vs cloud
What is on-premise document AI? OCR and document models on hardware you control
On-premise document AI explained: what it is, how it differs from cloud document APIs and SaaS tools, what the stack costs per page, and when to choose it.
Read the article