diff --git a/apps/site/docs/en/model-provider.mdx b/apps/site/docs/en/model-provider.mdx
index f0e5bd873..08d178a2b 100644
--- a/apps/site/docs/en/model-provider.mdx
+++ b/apps/site/docs/en/model-provider.mdx
@@ -50,6 +50,7 @@ Some advanced configs are also supported. Usually you don't need to use them.
 | `MIDSCENE_PREFERRED_LANGUAGE` | Optional. The preferred language for the model response. The default is `Chinese` if the current timezone is GMT+8 and `English` otherwise. |
 | `MIDSCENE_REPLANNING_CYCLE_LIMIT` | Optional. The maximum number of replanning cycles, default is 10 |
 | `OPENAI_MAX_TOKENS` | Optional. Maximum tokens for model response, default is 2048 |
+| `MIDSCENE_MODEL_TEMPERATURE` | Optional. Temperature used for model responses. Defaults to 0.1 (0.0 when using UI-TARS). |
 
 ### Debug configs
 
diff --git a/apps/site/docs/zh/model-provider.mdx b/apps/site/docs/zh/model-provider.mdx
index 279d88297..aceeb6723 100644
--- a/apps/site/docs/zh/model-provider.mdx
+++ b/apps/site/docs/zh/model-provider.mdx
@@ -53,6 +53,7 @@ Midscene 默认集成了 OpenAI SDK 调用 AI 服务。使用这个 SDK 限定
 | `MIDSCENE_PREFERRED_LANGUAGE` | 可选。模型响应的语言。如果当前时区是 GMT+8 则默认是 `Chinese`，否则是 `English` |
 | `MIDSCENE_REPLANNING_CYCLE_LIMIT` | 可选。最大重规划次数限制，默认是 10 |
 | `OPENAI_MAX_TOKENS` | 可选。模型响应的 max_tokens 数，默认是 2048 |
+| `MIDSCENE_MODEL_TEMPERATURE` | 可选。控制模型回复的 temperature，默认是 0.1（使用 UI-TARS 时为 0.0） |
 
 ### 调试配置
 
diff --git a/packages/core/src/ai-model/service-caller/index.ts b/packages/core/src/ai-model/service-caller/index.ts
index 3ebe12833..cd3259d7c 100644
--- a/packages/core/src/ai-model/service-caller/index.ts
+++ b/packages/core/src/ai-model/service-caller/index.ts
@@ -10,6 +10,7 @@ import {
   MIDSCENE_API_TYPE,
   MIDSCENE_LANGSMITH_DEBUG,
   OPENAI_MAX_TOKENS,
+  MIDSCENE_MODEL_TEMPERATURE,
   type TVlModeTypes,
   type UITarsModelVersion,
   globalConfigManager,
@@ -204,6 +205,8 @@ export async function callAI(
   const responseFormat = getResponseFormat(modelName, AIActionTypeValue);
 
   const maxTokens = globalConfigManager.getEnvConfigValue(OPENAI_MAX_TOKENS);
+  const temperatureConfig =
+    globalConfigManager.getEnvConfigValue(MIDSCENE_MODEL_TEMPERATURE);
   const debugCall = getDebug('ai:call');
   const debugProfileStats = getDebug('ai:profile:stats');
   const debugProfileDetail = getDebug('ai:profile:detail');
@@ -216,8 +219,16 @@ export async function callAI(
   let usage: OpenAI.CompletionUsage | undefined;
   let timeCost: number | undefined;
 
+  const defaultTemperature = vlMode === 'vlm-ui-tars' ? 0.0 : 0.1;
+  const parsedTemperature =
+    temperatureConfig && temperatureConfig.trim() !== ''
+      ? Number.parseFloat(temperatureConfig)
+      : undefined;
   const commonConfig = {
-    temperature: vlMode === 'vlm-ui-tars' ? 0.0 : 0.1,
+    temperature:
+      typeof parsedTemperature === 'number' && Number.isFinite(parsedTemperature)
+        ? parsedTemperature
+        : defaultTemperature,
     stream: !!isStreaming,
     max_tokens:
       typeof maxTokens === 'number'
diff --git a/packages/shared/src/env/types.ts b/packages/shared/src/env/types.ts
index 25792012a..7a1cc5b89 100644
--- a/packages/shared/src/env/types.ts
+++ b/packages/shared/src/env/types.ts
@@ -20,6 +20,7 @@ export const MIDSCENE_OPENAI_HTTP_PROXY = 'MIDSCENE_OPENAI_HTTP_PROXY';
 export const OPENAI_API_KEY = 'OPENAI_API_KEY';
 export const OPENAI_BASE_URL = 'OPENAI_BASE_URL';
 export const OPENAI_MAX_TOKENS = 'OPENAI_MAX_TOKENS';
+export const MIDSCENE_MODEL_TEMPERATURE = 'MIDSCENE_MODEL_TEMPERATURE';
 
 export const MIDSCENE_ADB_PATH = 'MIDSCENE_ADB_PATH';
 export const MIDSCENE_ADB_REMOTE_HOST = 'MIDSCENE_ADB_REMOTE_HOST';
@@ -204,6 +205,7 @@ export const NUMBER_ENV_KEYS = [
 
 export const STRING_ENV_KEYS = [
   OPENAI_MAX_TOKENS,
+  MIDSCENE_MODEL_TEMPERATURE,
   MIDSCENE_ADB_PATH,
   MIDSCENE_ADB_REMOTE_HOST,
   MIDSCENE_ADB_REMOTE_PORT,