FIX: nom du model

app.py

@@ -41,7 +41,7 @@ def trim_context(text: str, max_chars: int = 4000) -> str:
     # garder le début (head) utile pour les CSV preview
     return text[:max_chars] + "\n... (truncated)"
 
-def call_llm(messages, model="mistral-large", max_tokens=512, temperature=0.2):
+def call_llm(messages, model="mistral-small-latest", max_tokens=512, temperature=0.2):
     """Appel synchrones au LLM (OpenAI-compatible)."""
     resp = client.chat.completions.create(
         model=model,