Why Data Alchemy uses Claude AI (and now its own LLM too)

In Data Alchemy the LLM is the engine that reads and understands every document. The quality of that model determines the quality of the extraction. Here's how we choose the LLM, why we recommend Anthropic's Claude AI by default, and which other engines you can choose: GPT by OpenAI, Data Alchemy AI, Data Alchemy's proprietary model, and a compatible model through the European inference providers Scaleway and Regolo.ai.

A dedicated LLM for each document model

Data Alchemy doesn't orchestrate several models on the same document. It works differently, and more robustly: for each document model you create — invoices, delivery notes, orders, or any other type — the system pairs one LLM that handles all documents of that model.

This approach has concrete advantages:

• >Specialisation: each document type is handled by the model best suited to that specific task.
• >Consistency: all documents of the same model are read with the same logic, reducing variation in results.
• >Control: you know exactly which LLM processes each document type.

In other words, we don't pick a model "in general": we assign the right LLM to each document type.

Why we recommend Claude AI by default

For extracting data from documents and images, the engine we recommend by default is Anthropic's Claude AI. We use it every day on real business documents and value it in particular for:

• >Extraction accuracy: with Claude we state 99.8% accuracy on extracted data.
• >Instruction following: it returns exactly the requested fields, in the requested format.
• >Low tendency to hallucinate: when a value isn't there, it flags it instead of inventing one.
• >Understanding of documents and images: it reads scans, photos and complex layouts well.

That's why Claude AI is Data Alchemy's main engine and the one we propose to clients as the starting point, without preventing them from choosing another engine for an individual document model.

Data Alchemy AI and the other available engines

Alongside third-party engines we wanted to offer an alternative of our own: Data Alchemy AI, Data Alchemy's proprietary model, is available and runs on Data Alchemy's own servers and GPUs in Italy.

For each document model you choose one AI engine:

• >Claude (Anthropic), the engine we recommend by default;
• >GPT (OpenAI);
• >Data Alchemy AI, Data Alchemy's proprietary model, on its own servers and GPUs in Italy;
• >a compatible LLM through the European inference providers Scaleway (Generative APIs, with data centres in France) and Seeweb Regolo.ai (with data centres in Italy), on a subscription taken out directly by the customer.

Each document is processed by a single engine, the one assigned to its model: for example, you can use Claude for the invoice model, GPT for the delivery note model and Data Alchemy AI for the order model, or the same engine for all of them. Claude AI remains the engine we recommend by default.

What it means for your documents

For you, little changes in day-to-day use and a lot changes in substance: for every document type, you choose the engine you prefer. You can now choose between Claude, GPT, Data Alchemy AI or a compatible model through the European providers Scaleway and Regolo.ai, document model by document model.

---

Want to see the extraction quality on your own documents? Activate the free Demo (10 documents) and try it with 1 model chosen from the predefined ones, or contact us to plan a call with a demo on your own documents.