Expert knowledge for digital decisions

Which open language models are suitable for in-house use?

Short answer

Common models include Llama from Meta, Qwen from Alibaba, and Mistral from France. For German-language tasks, it is more important how well the model handles German than the model's name – this varies significantly and should be tested with your own texts, not rankings.

The Common Families

Family	Origin	Note
Llama	Meta	Widely used, many tools, own licensing terms
Qwen	Alibaba	Strong results, good multilingual capabilities
Mistral	France	European, efficient at small size
Gemma	Google	Compact, for simple tasks

Read Licenses Carefully

"Open" does not automatically mean "freely usable." Some models are under licenses with restrictions for commercial use or above a certain number of users. This should be checked before use – not afterwards.

How to Choose Correctly

Rankings on the internet say little about your use case. It makes sense to:

Collect twenty to fifty real cases from your daily life, with the correct answer for each.
Run two to three models with these cases.
Compare: hit rate, response time, costs.

This half day is worth more than any recommendation – even more than this one.

German as a Criterion

Many models are predominantly trained on English texts. For German technical texts, official language, or Swiss spelling, significant differences emerge. Test with your own texts, not with example sentences.

Important Note on Timeliness

Models appear and become outdated quickly. The mentioned families are a guideline, not a recommendation for a specific version. The decisive factor remains your own test with your own texts – it can be done in half a day and is more meaningful than any list.

Key facts

Common: Llama, Qwen, Mistral, Gemma
More important than the name: Test with your own German texts
Check before use: Licensing terms for commercial use