Mistral Small 3.2 — 24B
Snel, Europees, geschikt voor de meeste klantvragen.
via ZelixAI Privacy Cluster →
What is this model?
Mistral Small 3.2 is the 24-billion-parameter language model from Mistral AI in Paris. Open-weight (Apache 2.0 licence), instruction-tuned on a broad corpus with strong emphasis on European languages, it reaches 60–100 tokens per second on modern GPU hardware. We run this model inside the ZelixAI Privacy Cluster — your request never leaves the EU.
Strengths
Strengths: high inference speed, solid instruction-following for business conversations, reliable performance in French, German, Dutch, Italian and Spanish, and zero US-cloud components in the request path. For the majority of customer questions, this model produces results in our testing that do not lag behind the heavier cloud models — at significantly lower cost and with EU residency.
Best suited for
- General customer questions and chatbot conversations
- FAQ handling and knowledge-base queries
- Short summaries and intent detection
How ZelixAI uses this model
Within ZelixAI, Mistral Small is the default model for the Privacy Cluster — when you enable privacy mode (or when your package does so by default), we route to this model for customer questions, FAQ handling, short summaries and general chatbot conversations. You can switch to another model in the same tier or escalate to Cloud Premium at any time.
Real-world examples within ZelixAI
Concrete praktijkvoorbeelden voor dit model worden binnenkort hier gepubliceerd. Stel intussen vragen via onze contactpagina — we delen graag relevante use-cases uit onze klantbasis.
Limitations and caveats
Limitations: on highly specific questions without grounding context, the model can hallucinate. Not the first choice for complex code generation or multi-step reasoning — Llama 3.3 70B (Privacy) or Claude Sonnet 4 (Cloud) are stronger for that. Knowledge cut-off precedes the training date; always combine with our RAG layer (your own knowledge base) for up-to-date information.
Technical specifications
| Provider | ZelixAI Privacy Cluster |
| Context window | 128K tokens |
| Throughput | 40–100 tokens/s (Fast) |
| Cost tier | Very affordable |
| Tool / function-calling | yes |
| Data residency | EU (Netherlands · Germany · France) |