What task?
Sorting, extracting, summarising, short writing: an included model is usually enough. Reasoning through complex or ambiguous cases: that’s where a more powerful model can be worth it.
Guide
Which AI model for your agents?
The most powerful model isn’t always the best pick. To sort emails or fill in a quote, a simple model is often enough: you pay less, and it’s faster.
5 questions to ask yourself
What budget?
Each model adds (or doesn’t) a monthly add-on to your subscription. Start with an included model, and move up only if the results aren’t there.
Where does your data go?
With France / EU hosting, your data is processed in France or the EU. American models aren’t offered with this option.
How fast?
Bigger models often answer more slowly. For an agent that replies to your customers, speed matters as much as quality.
How much lock-in?
Open-weight models can be hosted by anyone. Closed models go through their vendor.
Our 3 model families
Chinese, included
Our default: open models, fast and cheap. More than enough for most repetitive tasks.
European
Models from Mistral AI, a French company. Pick them if you want a European vendor, with France / EU hosting.
American
The most powerful according to public leaderboards, and the most expensive. Often overkill for simple tasks: keep them for the cases that need them.
Models in the configurator
| Model | Vendor | Family | France / EU hosting | Add-on |
|---|---|---|---|---|
| DeepSeek V4 | DeepSeek | Chinese, included | Yes | Included |
| Qwen 3 | Alibaba | Chinese, included | Yes | Included |
| GLM-5 | Z.ai | Chinese, included | Yes | Included |
| Mistral Small 4 | Mistral AI | European | Yes | +€40 excl. VAT/month |
| Mistral Medium 3.5 | Mistral AI | European | Yes | +€90 excl. VAT/month |
| Mistral Large 3 | Mistral AI | European | Yes | +€150 excl. VAT/month |
| Gemini Pro | American | No | +€190 excl. VAT/month | |
| Claude Sonnet 5.5 | Anthropic | American | No | +€190 excl. VAT/month |
| GPT-6 | OpenAI | American | No | +€220 excl. VAT/month |
| Claude Opus 5.5 | Anthropic | American | No | +€290 excl. VAT/month |
DeepSeek V4
Qwen 3
GLM-5
Mistral Small 4
Gemini Pro
Claude Sonnet 5.5
GPT-6Where to read the benchmarks
We don’t copy scores here: they change every week. These are the public leaderboards we check, and what each one measures.
- Artificial Analysis : An index that combines several tests (reasoning, code, maths), plus each model’s price and speed.
- Arena (formerly LMArena) : Blind human votes between two answers: useful to judge perceived quality in real use.
- SWE-bench : The ability to fix real bugs in real projects: the reference for code.
As of October 2026, the latest American models sit at the top of these leaderboards, with several open Chinese models close behind. Exact version names vary between leaderboards: always check the date of the snapshot.
Vendor sources
- Mistral AI : Official site
- DeepSeek : Models published on Hugging Face
- Qwen (Alibaba) : Models published on Hugging Face
- GLM (Z.ai) : Models published on Hugging Face
- Anthropic : Official site
- OpenAI : Official site
- Google Gemini : Official documentation
Model and vendor names are trademarks of their respective owners. No partnership or affiliation.
Logos: files from Wikimedia Commons, unmodified (DeepSeek icon under the MIT licence; the others public domain or CC0). See on Wikimedia Commons