diff --git a/src/renderer/src/aiCore/clients/anthropic/AnthropicAPIClient.ts b/src/renderer/src/aiCore/clients/anthropic/AnthropicAPIClient.ts
index 73a5bb61c..e18b20889 100644
--- a/src/renderer/src/aiCore/clients/anthropic/AnthropicAPIClient.ts
+++ b/src/renderer/src/aiCore/clients/anthropic/AnthropicAPIClient.ts
@@ -524,9 +524,18 @@ export class AnthropicAPIClient extends BaseApiClient<
           switch (rawChunk.type) {
             case 'message': {
               let i = 0
+              let hasTextContent = false
+              let hasThinkingContent = false
+
               for (const content of rawChunk.content) {
                 switch (content.type) {
                   case 'text': {
+                    if (!hasTextContent) {
+                      controller.enqueue({
+                        type: ChunkType.TEXT_START
+                      } as TextStartChunk)
+                      hasTextContent = true
+                    }
                     controller.enqueue({
                       type: ChunkType.TEXT_DELTA,
                       text: content.text
@@ -539,6 +548,12 @@ export class AnthropicAPIClient extends BaseApiClient<
                     break
                   }
                   case 'thinking': {
+                    if (!hasThinkingContent) {
+                      controller.enqueue({
+                        type: ChunkType.THINKING_START
+                      } as ThinkingStartChunk)
+                      hasThinkingContent = true
+                    }
                     controller.enqueue({
                       type: ChunkType.THINKING_DELTA,
                       text: content.thinking
diff --git a/src/renderer/src/aiCore/clients/openai/OpenAIApiClient.ts b/src/renderer/src/aiCore/clients/openai/OpenAIApiClient.ts
index b08f179fb..393e8733d 100644
--- a/src/renderer/src/aiCore/clients/openai/OpenAIApiClient.ts
+++ b/src/renderer/src/aiCore/clients/openai/OpenAIApiClient.ts
@@ -5,6 +5,7 @@ import {
   GEMINI_FLASH_MODEL_REGEX,
   getOpenAIWebSearchParams,
   isDoubaoThinkingAutoModel,
+  isQwenReasoningModel,
   isReasoningModel,
   isSupportedReasoningEffortGrokModel,
   isSupportedReasoningEffortModel,
@@ -166,10 +167,17 @@ export class OpenAIAPIClient extends OpenAIBaseClient<
 
     // Qwen models
     if (isSupportedThinkingTokenQwenModel(model)) {
-      return {
+      const thinkConfig = {
         enable_thinking: true,
         thinking_budget: budgetTokens
       }
+      if (this.provider.id === 'dashscope') {
+        return {
+          ...thinkConfig,
+          incremental_output: true
+        }
+      }
+      return thinkConfig
     }
 
     // Grok models
@@ -436,7 +444,14 @@ export class OpenAIAPIClient extends OpenAIBaseClient<
         messages: OpenAISdkMessageParam[]
         metadata: Record<string, any>
       }> => {
-        const { messages, mcpTools, maxTokens, streamOutput, enableWebSearch } = coreRequest
+        const { messages, mcpTools, maxTokens, enableWebSearch } = coreRequest
+        let { streamOutput } = coreRequest
+
+        // Qwen3商业版（思考模式）、Qwen3开源版、QwQ、QVQ只支持流式输出。
+        if (isQwenReasoningModel(model)) {
+          streamOutput = true
+        }
+
         // 1. 处理系统消息
         let systemMessage = { role: 'system', content: assistant.prompt || '' }
 
diff --git a/src/renderer/src/store/thunk/messageThunk.ts b/src/renderer/src/store/thunk/messageThunk.ts
index 1cd98aa31..ad2e63e55 100644
--- a/src/renderer/src/store/thunk/messageThunk.ts
+++ b/src/renderer/src/store/thunk/messageThunk.ts
@@ -446,7 +446,7 @@ const fetchAndProcessAssistantResponseImpl = async (
             content: accumulatedContent,
             status: MessageBlockStatus.STREAMING
           }
-          smartBlockUpdate(initialPlaceholderBlockId, changes, MessageBlockType.MAIN_TEXT)
+          smartBlockUpdate(initialPlaceholderBlockId, changes, MessageBlockType.MAIN_TEXT, true)
           mainTextBlockId = initialPlaceholderBlockId
           initialPlaceholderBlockId = null
         } else if (!mainTextBlockId) {
@@ -496,7 +496,7 @@ const fetchAndProcessAssistantResponseImpl = async (
           }
           thinkingBlockId = initialPlaceholderBlockId
           initialPlaceholderBlockId = null
-          smartBlockUpdate(thinkingBlockId, changes, MessageBlockType.THINKING)
+          smartBlockUpdate(thinkingBlockId, changes, MessageBlockType.THINKING, true)
         } else if (!thinkingBlockId) {
           const newBlock = createThinkingBlock(assistantMsgId, accumulatedThinking, {
             status: MessageBlockStatus.STREAMING,
diff --git a/src/renderer/src/types/sdk.ts b/src/renderer/src/types/sdk.ts
index ae7d823d4..1eaf74d86 100644
--- a/src/renderer/src/types/sdk.ts
+++ b/src/renderer/src/types/sdk.ts
@@ -54,6 +54,7 @@ export type ReasoningEffortOptionalParams = {
   reasoning_effort?: OpenAI.Chat.Completions.ChatCompletionCreateParams['reasoning_effort'] | 'none' | 'auto'
   enable_thinking?: boolean
   thinking_budget?: number
+  incremental_output?: boolean
   enable_reasoning?: boolean
   extra_body?: Record<string, any>
   // Add any other potential reasoning-related keys here if they exist