The short answer
CALC does not depend on one fixed, off-the-shelf language model. It uses a modular multi-model architecture designed for structural calculations and design checks. Depending on the task and the reasoning tier you choose, a suitable model from one of several contracted AI providers handles your request. A language model on its own quickly reaches its limits with formulas, references to standards, and reliable numbers. That is why Mia calculates numbers with program code, takes standard values from the Dlubal Knowledge Base, and follows fixed steps in structural Skills.
Why one language model is not enough
General language models write fluent text, but they can miscalculate a value or cite a clause that does not exist. For structural analysis, that is not reliable enough.
In CALC, the language model understands your question and plans the steps. The numbers are calculated in program code, standard values come from the Knowledge Base, and a Skill sets the order of input, calculation, design check, and PDF report.
How CALC selects a model
Each request is handled by one model. The reasoning tier you choose in the chat, Standard or High, determines which model is used. If a provider is unavailable, a fallback provider takes over. The models do not communicate with each other.
We test new models before we use them and switch when another model works more reliably. That is why this page does not name a fixed model.
Data protection and responsibility
Which AI providers process your requests and where processing can take place is explained in a separate FAQ. As with any structural analysis software, you remain responsible for checking the results.