diff --git a/src/gemini_utils.py b/src/gemini_utils.py
index ab640b18..3d6643fa 100644
--- a/src/gemini_utils.py
+++ b/src/gemini_utils.py
@@ -293,9 +293,13 @@ async def call_gemini_with_retry(prompt: str, response_format: str = "json", max
     consecutive_429s = 0
     base_wait_time = 2.0
     
-    # 1. Smart Re-ordering
+    # 1. Smart Filtering and Re-ordering (Strict Lite-Only Policy for Tier 1)
     if prefer_flash:
-        models = sorted(models_pool, key=lambda m: 0 if "flash" in m or "lite" in m else 1)
+        # Strict filter: Only allow flash/lite models, maintaining version sorting from pool
+        models = [m for m in models_pool if "flash" in m or "lite" in m]
+        if not models:
+            # Emergency fallback: if no flash/lite discovered, use hardcoded safe defaults
+            models = ["gemini-1.5-flash", "gemini-1.5-flash-latest"]
     else:
         models = models_pool
 
diff --git a/src/v2_optimizer.py b/src/v2_optimizer.py
index 3f249383..042ee648 100644
--- a/src/v2_optimizer.py
+++ b/src/v2_optimizer.py
@@ -319,7 +319,7 @@ class V2VisionEngine:
             analyst_results = []
 
             # 1.1 Fast-Track: Large Batches, NO GROUNDING (Fast)
-            BATCH_SIZE_FAST = 100 # Increased from 40 for optimal RPM/TPM balance
+            BATCH_SIZE_FAST = 50 # Balanced "Sweet Spot" for RPM/TPM and timeout safety (2026)
             total_fast = len(fast_track)
             for i in range(0, total_fast, BATCH_SIZE_FAST):
                 batch = fast_track[i:i+BATCH_SIZE_FAST]