diff --git a/config.yaml b/config.yaml
index 18f9d1d..84d2048 100644
--- a/config.yaml
+++ b/config.yaml
@@ -19,7 +19,7 @@ ai:
   timeout_seconds: 30
   voter_a:
     provider: gemini       # gemini | claude | groq | deepseek | xai | openrouter | ollama
-    model: gemini-2.5-flash
+    model: gemini-2.5-flash-lite
     temperature: 0.1
   voter_b:
     provider: groq
diff --git a/src/aitrader/ai/gemini.py b/src/aitrader/ai/gemini.py
index f3aa871..5bed728 100644
--- a/src/aitrader/ai/gemini.py
+++ b/src/aitrader/ai/gemini.py
@@ -2,8 +2,11 @@
 from __future__ import annotations
 
 import json
+import re
+import time
 
 import google.generativeai as genai
+from google.api_core.exceptions import ResourceExhausted
 
 from ..config import Settings
 from ..logging_setup import get_logger
@@ -13,6 +16,11 @@ from .schema import JSON_SCHEMA, TradeDecision
 log = get_logger(__name__)
 
 
+def _extract_retry_seconds(err: Exception, fallback: float) -> float:
+    m = re.search(r"retry in (\d+(?:\.\d+)?)", str(err))
+    return float(m.group(1)) if m else fallback
+
+
 class GeminiClient:
     provider = "gemini"
 
@@ -31,15 +39,31 @@ class GeminiClient:
         self.timeout = settings.ai.timeout_seconds
 
     def decide(self, user_prompt: str) -> TradeDecision:
-        resp = self._model.generate_content(
-            user_prompt,
-            generation_config={
-                "response_mime_type": "application/json",
-                "response_schema": JSON_SCHEMA,
-                "temperature": self.temperature,
-            },
-            request_options={"timeout": self.timeout},
-        )
+        attempts = 0
+        max_attempts = 3
+        while True:
+            try:
+                resp = self._model.generate_content(
+                    user_prompt,
+                    generation_config={
+                        "response_mime_type": "application/json",
+                        "response_schema": JSON_SCHEMA,
+                        "temperature": self.temperature,
+                    },
+                    request_options={"timeout": self.timeout},
+                )
+                break
+            except ResourceExhausted as e:
+                attempts += 1
+                if attempts >= max_attempts:
+                    log.warning("gemini.rate_limit_giveup", attempts=attempts)
+                    return TradeDecision(
+                        action="HOLD", confidence=0.0, suggested_size_pct=0.0,
+                        reasoning="rate_limit_exhausted",
+                    )
+                wait = min(_extract_retry_seconds(e, 30.0) + 2, 60.0)
+                log.warning("gemini.rate_limit", attempt=attempts, wait_s=wait)
+                time.sleep(wait)
         text = resp.text or "{}"
         try:
             data = json.loads(text)