diff --git a/app/services/llm_client.py b/app/services/llm_client.py
index e6f2a79..1346400 100644
--- a/app/services/llm_client.py
+++ b/app/services/llm_client.py
@@ -239,6 +239,24 @@ class LLMClient:
     def _endpoint(self) -> str:
         return f"{self.api_base.rstrip('/')}/chat/completions"
 
+    @staticmethod
+    def _http_error_detail(exc: httpx.HTTPStatusError) -> str:
+        """Human-readable HTTP error including the provider's response body.
+
+        Providers return actionable JSON on 4xx (e.g. « model not allowed »,
+        « country not supported »); surfacing it makes the Settings test
+        debuggable instead of a bare « 403 Forbidden ».
+        """
+        resp = exc.response
+        try:
+            body = (resp.text or "").strip()
+        except Exception:  # noqa: BLE001 — body already consumed / undecodable
+            body = ""
+        if len(body) > 500:
+            body = body[:500] + "…"
+        base = f"{resp.status_code} {resp.reason_phrase} ({resp.url})"
+        return f"{base}: {body}" if body else base
+
     async def _http_complete(self, messages, model, tools, *, _noticer: str = "") -> LLMResponse:
         payload: dict = {
             "model": model,
@@ -272,7 +290,8 @@ class LLMClient:
                     f"Le modèle « {model} » n'est plus disponible ({exc.response.status_code}). "
                     f"Réponse générée avec « {self.default_model} » à la place."
                 ))
-            raise
+            # Surface the provider's own error body (403 « forbidden », 400 …).
+            raise RuntimeError(self._http_error_detail(exc)) from exc
 
         response = self._parse_response(data, model)
         response.notice = _noticer or ""
diff --git a/app/services/llm_config.py b/app/services/llm_config.py
index ce0f0a5..cbba77a 100644
--- a/app/services/llm_config.py
+++ b/app/services/llm_config.py
@@ -347,7 +347,13 @@ async def fetch_provider_models(provider: str, *, api_key: str = "",
 
     async with httpx.AsyncClient(timeout=timeout) as client:
         resp = await client.get(f"{base_url}/models", headers=headers)
-        resp.raise_for_status()
+        if resp.status_code >= 400:
+            body = (resp.text or "").strip()
+            if len(body) > 500:
+                body = body[:500] + "…"
+            raise RuntimeError(
+                f"{resp.status_code} {resp.reason_phrase} ({resp.url}): {body}"
+            )
         data = resp.json()
 
     ids: list[str] = []