diff --git a/apps/worker/worker_app/tasks/vision/fast_path.py b/apps/worker/worker_app/tasks/vision/fast_path.py
index f8d3435d8..4ec8d33c3 100644
--- a/apps/worker/worker_app/tasks/vision/fast_path.py
+++ b/apps/worker/worker_app/tasks/vision/fast_path.py
@@ -23,10 +23,10 @@ logger = logging.getLogger(__name__)
 
 # 超时（可通过环境变量覆盖）
 _IMG_WORKERS = int(os.environ.get("VISION_V2_IMG_WORKERS", "8"))
-_FAST_TIMEOUT = float(os.environ.get("VISION_V2_FAST_TIMEOUT", "8"))
-_FAST_JSON_TIMEOUT = float(os.environ.get("VISION_V2_FAST_JSON_TIMEOUT", "8"))
+_FAST_TIMEOUT = float(os.environ.get("VISION_V2_FAST_TIMEOUT", "12"))
+_FAST_JSON_TIMEOUT = float(os.environ.get("VISION_V2_FAST_JSON_TIMEOUT", "12"))
 _OCR_TIMEOUT = float(os.environ.get("VISION_V2_OCR_TIMEOUT", "6"))
-_PRO_TIMEOUT = float(os.environ.get("VISION_V2_PRO_TIMEOUT", "20"))
+_PRO_TIMEOUT = float(os.environ.get("VISION_V2_PRO_TIMEOUT", "25"))
 
 _FALLBACK_RESULT = {
     "name": "未识别",
diff --git a/apps/worker/worker_app/tasks/vision/vlm_fallback.py b/apps/worker/worker_app/tasks/vision/vlm_fallback.py
index c9b9e1d61..ae8276535 100644
--- a/apps/worker/worker_app/tasks/vision/vlm_fallback.py
+++ b/apps/worker/worker_app/tasks/vision/vlm_fallback.py
@@ -17,7 +17,7 @@ logger = logging.getLogger(__name__)
 
 _BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
 _PRO_MODEL = "qwen3.7-plus"
-_DEFAULT_TIMEOUT = 20
+_DEFAULT_TIMEOUT = 25
 _DEFAULT_MAX_TOKENS = 800
 
 _PRO_SYSTEM = (
@@ -113,6 +113,7 @@ def call_pro_vlm(
         "max_tokens": _DEFAULT_MAX_TOKENS,
         "stream": False,
         "enable_thinking": False,
+        "response_format": {"type": "json_object"},
     }
     try:
         r = httpx.post(
diff --git a/apps/worker/worker_app/tasks/vision/vlm_fast_json.py b/apps/worker/worker_app/tasks/vision/vlm_fast_json.py
index 3de9bba1b..8ad44400f 100644
--- a/apps/worker/worker_app/tasks/vision/vlm_fast_json.py
+++ b/apps/worker/worker_app/tasks/vision/vlm_fast_json.py
@@ -7,7 +7,7 @@
 - enable_thinking=false 关闭推理链（reasoning 是延迟主因）
 - system prompt 极致精简，只给字段 schema 和强约束（禁止自然语言、禁止 markdown）
 - max_tokens=350、temperature=0.1（稳定输出 JSON）
-- timeout=8s（失败由外层走 pro 兜底）
+- timeout=12s（失败由外层走 pro 兜底）
 - API Key 从环境变量 DASHSCOPE_API_KEY 读取
 """
 
@@ -24,7 +24,7 @@ logger = logging.getLogger(__name__)
 # DashScope OpenAI 兼容 endpoint
 _BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
 _FAST_MODEL = "qwen3.8-flash"
-_DEFAULT_TIMEOUT = 8
+_DEFAULT_TIMEOUT = 12
 _DEFAULT_MAX_TOKENS = 350
 
 # 极简 system prompt：只给字段定义 + 硬性输出要求
@@ -108,6 +108,7 @@ def call_fast_json(
         "max_tokens": max_tokens,
         "stream": False,
         "enable_thinking": False,
+        "response_format": {"type": "json_object"},
     }
     try:
         resp = httpx.post(