diff --git a/src/config/api_ada_configs.py b/src/config/api_ada_configs.py
index 24b7896e..a5d5f059 100644
--- a/src/config/api_ada_configs.py
+++ b/src/config/api_ada_configs.py
@@ -63,6 +63,9 @@ class ModelInfo(ConfigBase):
     temperature: float | None = field(default=None)
     """模型级别温度（可选），会覆盖任务配置中的温度"""
 
+    max_tokens: int | None = field(default=None)
+    """模型级别最大token数（可选），会覆盖任务配置中的max_tokens"""
+
     force_stream_mode: bool = field(default=False)
     """是否强制使用流式输出模式"""
 
diff --git a/src/llm_models/utils_model.py b/src/llm_models/utils_model.py
index 2d2c1c86..7fb4cfd0 100644
--- a/src/llm_models/utils_model.py
+++ b/src/llm_models/utils_model.py
@@ -324,11 +324,20 @@ class LLMRequest:
                     if effective_temperature is None:
                         effective_temperature = self.model_for_task.temperature
                     
+                    # max_tokens 优先级：参数传入 > 模型级别配置 > extra_params > 任务配置
+                    effective_max_tokens = max_tokens
+                    if effective_max_tokens is None:
+                        effective_max_tokens = model_info.max_tokens
+                    if effective_max_tokens is None:
+                        effective_max_tokens = (model_info.extra_params or {}).get("max_tokens")
+                    if effective_max_tokens is None:
+                        effective_max_tokens = self.model_for_task.max_tokens
+                    
                     return await client.get_response(
                         model_info=model_info,
                         message_list=(compressed_messages or message_list),
                         tool_options=tool_options,
-                        max_tokens=self.model_for_task.max_tokens if max_tokens is None else max_tokens,
+                        max_tokens=effective_max_tokens,
                         temperature=effective_temperature,
                         response_format=response_format,
                         stream_response_handler=stream_response_handler,
diff --git a/template/model_config_template.toml b/template/model_config_template.toml
index 1e072d13..225b94f1 100644
--- a/template/model_config_template.toml
+++ b/template/model_config_template.toml
@@ -1,5 +1,5 @@
 [inner]
-version = "1.8.2"
+version = "1.9.0"
 
 # 配置文件版本号迭代规则同bot_config.toml
 
@@ -54,9 +54,11 @@ name = "siliconflow-deepseek-v3.2"
 api_provider = "SiliconFlow"
 price_in = 2.0
 price_out = 3.0
+# temperature = 0.5    # 可选：为该模型单独指定温度，会覆盖任务配置中的温度
+# max_tokens = 4096    # 可选：为该模型单独指定最大token数，会覆盖任务配置中的max_tokens
 [models.extra_params] # 可选的额外参数配置
 enable_thinking = false # 不启用思考
-# temperature = 0.5    # 可选：为该模型单独指定温度，会覆盖任务配置中的温度
+
 
 [[models]]
 model_identifier = "deepseek-ai/DeepSeek-V3.2-Exp"
@@ -64,9 +66,11 @@ name = "siliconflow-deepseek-v3.2-think"
 api_provider = "SiliconFlow"    
 price_in = 2.0
 price_out = 3.0
+# temperature = 0.7    # 可选：为该模型单独指定温度，会覆盖任务配置中的温度
+# max_tokens = 4096    # 可选：为该模型单独指定最大token数，会覆盖任务配置中的max_tokens
 [models.extra_params] # 可选的额外参数配置
 enable_thinking = true # 启用思考
-# temperature = 0.7    # 可选：为该模型单独指定温度，会覆盖任务配置中的温度
+
 
 [[models]]
 model_identifier = "Qwen/Qwen3-Next-80B-A3B-Instruct"