Update tools.py

tools.py

@@ -219,7 +219,7 @@ def answer_image_question(task_id: str = "", file_path: str = "", question: str
         """.strip()
 
         response = client.chat.completions.create(
-            model=os.getenv("DASHSCOPE_VL_MODEL", "qwen-vl-plus-latest"),
+            model=os.getenv("DASHSCOPE_VL_MODEL", "qwen3.6-plus"),
             messages=[
                 {
                     "role": "user",
@@ -647,7 +647,7 @@ def _extract_final_answer_from_transcript(
     This is intentionally separate from transcription to reduce hallucination.
     """
     client = _dashscope_client()
-    text_model = os.getenv("DASHSCOPE_TEXT_MODEL", "qwen-plus")
+    text_model = os.getenv("DASHSCOPE_TEXT_MODEL", "qwen3.5-flash")
 
     prompt = f"""
 You are an exact-match answer extractor for an evaluation benchmark.