fix: heavy tier — Dense first for reasoning, MoE workhorse, VLM overflow

router/router.py

@@ -119,7 +119,11 @@ def is_gpu_busy(model):
     return active >= max_c
 
 def select_best_gpu(candidates, reason):
-    """Pick the best GPU from candidates, preferring least-loaded."""
+    """Pick the best GPU from candidates IN ORDER — first non-busy one wins."""
+    for m in candidates:
+        if not is_gpu_busy(m):
+            return {"model": m, "reason": reason}
+    # All busy — pick least loaded
     best = None
     best_load = 999
     for m in candidates:
@@ -128,10 +132,7 @@ def select_best_gpu(candidates, reason):
             best_load = load
             best = m
     if best:
-        actual_reason = reason
+        return {"model": best, "reason": "load_balanced_" + reason}
-        if is_gpu_busy(best):
-            actual_reason = "load_balanced_" + reason
-        return {"model": best, "reason": actual_reason}
     return None
 
 def route(rd, tier):
@@ -182,7 +183,7 @@ def route(rd, tier):
     
     # TIER 3: Heavy reasoning — large context or very long conversations
     if t > 4000 or turns > 8:
-        candidates = [m for m in ["qwen3.6-35B-A3B","qwen3.5-9b-vlm","qwen3.6-27B-code"] if m in avail]
+        candidates = [m for m in ["qwen3.6-27B-code","qwen3.6-35B-A3B","qwen3.5-9b-vlm"] if m in avail]
         result = select_best_gpu(candidates, "heavy_reasoning")
         if result: return result