ZelixAI is a meta-platform — you choose per bot which AI model to use. Available providers:
- Anthropic Claude — Opus 4.7, Sonnet 4.6, Haiku 4.5 (best for tone of voice)
- OpenAI GPT — GPT-5, GPT-4.1, GPT-4o (industry standard)
- Google Gemini — Pro, Flash (best in cost-per-token)
- ★ ZelixAI Best Choice Mix — Mistral via OVH — Large, Small (fully EU-sovereign, NL-hosted) — our recommended default for EU compliance
- ★ ZelixAI Best Choice Mix — OpenRouter — 369 alternative models (Llama, DeepSeek, Qwen, Mistral-open) — our recommended routing for the widest choice + best price-performance
- Local LLM model — hosted on your own hardware or at our data center, for privacy mode and strict data residency
Our "Best Choice Mix" combines OVH (EU-sovereign) with OpenRouter (widest choice) — giving you automatically the right model per bot for the right task. Switch manually or let ZelixAI's router handle it — in 1 click.
Best-Choice Routing: ZelixAI advises per use case
ZelixAI analyses your use case during bot configuration and recommends which model best fits. You are free to choose — the advice is indicative.
| Use case |
Recommended model |
Reason |
| NL customer service (simple questions) |
Mistral via OVH |
EU-sovereign, NL-hosted, excellent NL comprehension |
| Multilingual customer service |
GPT-4o-mini |
Strongest multilingual performance, low cost |
| Complex Q&A / legal / technical |
Claude Sonnet |
Best reasoning and tone-of-voice quality |
| High-volume simple FAQ |
Claude Haiku / Gemini Flash |
Lowest cost per token, fast |
| Privacy-first / no external calls |
Local LLM |
Own hardware or our data center, no external API |
Model evaluation framework
Every model in our stack is continuously benchmarked on 5 dimensions: (1) language comprehension NL/EN/DE/FR, (2) tone-of-voice preservation, (3) factual accuracy on knowledge base content, (4) refusal rate on out-of-scope questions, (5) latency p95. Results drive our Best-Choice-Routing recommendations.