diff --git a/backend/app/services/gemini.py b/backend/app/services/gemini.py
index f189b89..1ae0c88 100644
--- a/backend/app/services/gemini.py
+++ b/backend/app/services/gemini.py
@@ -54,7 +54,7 @@ class GeminiService:
         self.prompts_dir = Path(__file__).parent.parent / "prompts"
 
     async def _generate(self, contents: Any, config: Any = None) -> tuple[Any, str]:
-        """Call generate_content, falling back on 429/quota errors. Returns (response, model_used)."""
+        """Call generate_content, falling back on 429/503 transient errors. Returns (response, model_used)."""
         for model in [self.model_name, *self._fallback_models]:
             try:
                 kw: dict[str, Any] = {"model": model, "contents": contents}
@@ -62,12 +62,12 @@ class GeminiService:
                     kw["config"] = config
                 response = await asyncio.to_thread(client.models.generate_content, **kw)
                 if model != self.model_name:
-                    logger.warning(f"Used fallback model {model!r} (primary quota exceeded)")
+                    logger.warning(f"Used fallback model {model!r} (primary unavailable)")
                 return response, model
             except Exception as exc:
                 msg = str(exc)
-                if "429" in msg or "RESOURCE_EXHAUSTED" in msg:
-                    logger.warning(f"Model {model!r} quota exceeded, trying next fallback")
+                if "429" in msg or "RESOURCE_EXHAUSTED" in msg or "503" in msg or "UNAVAILABLE" in msg:
+                    logger.warning(f"Model {model!r} unavailable, trying next fallback")
                     last_exc: Exception = exc
                     continue
                 raise