How Auto picks the model
Auto reads your request, estimates how demanding it is, then takes the model that is enough within your plan. The reason is written under every answer.
1. The kind of request
Writing, analysis, translation, reasoning, code, creative work or a simple exchange. An attached document counts.
2. How demanding it is
Short, standard or demanding. A demanding request moves up a notch, to an advanced model.
3. What your plan allows
Daily quota reached or model unavailable: the best fast model takes over, and the reason says so.
The routing table
Auto's first choice for each kind of request, as the app reads it. If that model is unavailable, the next one in the same row takes its place.
| Request | Standard, first choice | Demanding, first choice |
|---|---|---|
| Writing | Claude Haiku 5.5 | Claude Sonnet 5.5 |
| Analysis | Gemini 3.8 Flash | Gemini 4 Argon |
| Translation | Mistral Small 4 | Mistral Large 4 |
| Reasoning | Gemini 3.8 Flash | GPT-6.1 Sol |
| Code | Claude Haiku 5.5 | Claude Sonnet 5.5 |
| Creative writing | Claude Haiku 5.5 | Claude Sonnet 5.5 |
| Simple exchange | Mistral Small 4 | Gemini 3.8 Flash |
Grok 4.7, Claude Opus 5.5, GPT-6 Astra are never picked by Auto: you choose them by hand (expert models from Pro).
When the rules hesitate between two kinds, the app also asks a very lean classifier; the choice may then differ from the one shown here.
Try it on your question
Type a request from your work: Auto says which model it would take, and why. No answer is generated here.
Gemini 3.8 Flash
Short analysis: Gemini 3.8 Flash, very fast, handles long documents, without paying for a bigger model than needed.
Free messages with no card, valid for 30 days, once per person.
When a daily quota is reached
Auto moves to the best model still available in your plan and says so in its reason. Nothing extra is billed: quotas are daily, never unlimited, with no surcharge.
On video
“Guess the model”: four requests, and the model Auto picks for each. 30 seconds, in French.
