Skip to content

Pick your style: Denkbaar

Book a conversation

Articles

Knowledge

Claude for making, Mistral on your own server for fixed processes

Making and trying something quickly needs a different model than a business process that must run for years. This is how we choose.

Two people review a design beside a laptop and a server.
Building and long-term operation call for different choices. Illustration: Denkbaar.

We use two kinds of models for two kinds of work. To make and try things: Claude, now Opus 5.5. For fixed business processes that must run for a long time: our own software on our own server, with an open model such as Mistral.

Plain technical English A model through an API runs at the vendor; an open model can run on your own server. We use Claude Opus 5.5 to make software. We put an open model such as Mistral or Llama in the software that keeps running.

Why it matters

Our advice to customers: go fast, make and try things: Claude. Automate fixed business processes for long-term use: your own software on your own server, with Mistral as the AI. About Mistral: no 100% guarantee, but more certainty than Claude or ChatGPT, and less dependent on the US.

A process that runs for years should not depend on the price, the rules or the availability of one vendor.

A laptop shows separate design blocks beside a few cards.
Use a strong model to build and try things quickly. Illustration: Denkbaar.

How it works

noyes

Idea

Make and try with Claude

Does it stay?

Throw away, lesson learned

Own software on own server

Open model: Mistral or Llama

Make it reusable

You make it fast with Claude; what stays becomes your own software with an open model on your own server.

  • Making: Claude Opus 5.5 came out on 22 September 2026. In our own desk research at the end of October 2026, Opus 5.5 scored highest on the leaderboards we checked. We did not run those benchmarks ourselves; we read the public leaderboards.
  • Running: Mistral publishes the licence of each model. Mistral Small 4 and the Ministral 3 models are under Apache 2.0. Llama has its own licence from Meta, with conditions.
A server stands beside a chip and tidy document trays.
Test an open model on your own examples. Illustration: Denkbaar.

What you do

  1. Make your first version with the strongest model you have.
  2. Does it become a fixed process? Put the steps in code, and use AI only to read and write text.
  3. For that step, choose an open model with a licence that allows business use.
  4. Test the open model on your own examples before you switch.

What we can do with you: we make your idea quickly with Claude, and put what stays on your own server with an open model. Book a conversation.

Sources

  1. Claude release notes (Claude Opus 5.5, 2026-09-22)
  2. Mistral Docs: Models (licences per model)
  3. Meta: Llama 3.1 Community License Agreement

Topics: modelsmistralclaudehosting