From b095a5511aac9fc9ca9d0734dc5c07c5c5772106 Mon Sep 17 00:00:00 2001
From: euvre <93761161+euvre@users.noreply.github.com>
Date: Thu, 23 Jul 2026 11:14:08 +0800
Subject: [PATCH] fix: make qwen3-vl-plus usable as chat/vision model in Go
 mode (#17254)

---
 conf/models/aliyun.json               | 10 +++++++---
 conf/models/qiniu.json                |  2 +-
 internal/entity/models/aliyun.go      | 25 +++++++++++++++++++++++++
 web/src/hooks/use-document-request.ts |  1 -
 4 files changed, 33 insertions(+), 5 deletions(-)

diff --git a/conf/models/aliyun.json b/conf/models/aliyun.json
index 7511f28c38..d003907656 100644
--- a/conf/models/aliyun.json
+++ b/conf/models/aliyun.json
@@ -32,10 +32,14 @@
     },
     {
       "name": "qwen3-vl-plus",
-      "max_tokens": 260096,
+      "max_tokens": 256000,
       "model_types": [
-        "vision"
-      ]
+        "vision",
+        "chat"
+      ],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen-tts-flash",
diff --git a/conf/models/qiniu.json b/conf/models/qiniu.json
index ab4f2b485b..0e0b18eac5 100644
--- a/conf/models/qiniu.json
+++ b/conf/models/qiniu.json
@@ -551,4 +551,4 @@
       }
     }
   ]
-}
\ No newline at end of file
+}
diff --git a/internal/entity/models/aliyun.go b/internal/entity/models/aliyun.go
index 80292015ed..6dd2cbb7f2 100644
--- a/internal/entity/models/aliyun.go
+++ b/internal/entity/models/aliyun.go
@@ -112,6 +112,11 @@ func (a *AliyunModel) ChatWithMessages(ctx context.Context, modelName string, me
 		}
 	}
 
+	// For qwen3 models on DashScope, enable_thinking defaults to true when
+	// omitted. RAGFlow's default is to disable thinking unless explicitly
+	// enabled by the user, matching Python's chat_model.py behavior.
+	applyQwen3ThinkingDefault(modelName, reqBody)
+
 	jsonData, err := json.Marshal(reqBody)
 	if err != nil {
 		return nil, fmt.Errorf("failed to marshal request: %w", err)
@@ -253,6 +258,11 @@ func (a *AliyunModel) ChatStreamlyWithSender(ctx context.Context, modelName stri
 		}
 	}
 
+	// For qwen3 models on DashScope, enable_thinking defaults to true when
+	// omitted. RAGFlow's default is to disable thinking unless explicitly
+	// enabled by the user, matching Python's chat_model.py behavior.
+	applyQwen3ThinkingDefault(modelName, reqBody)
+
 	jsonData, err := json.Marshal(reqBody)
 	if err != nil {
 		return fmt.Errorf("failed to marshal request: %w", err)
@@ -337,6 +347,21 @@ func (a *AliyunModel) ChatStreamlyWithSender(ctx context.Context, modelName stri
 	return sender(&endOfStream, nil)
 }
 
+// applyQwen3ThinkingDefault ensures enable_thinking=false is sent for qwen3
+// models when it hasn't been explicitly configured. DashScope defaults
+// enable_thinking to true for qwen3 models, which produces reasoning output
+// that RAGFlow doesn't expect in most pipelines. Mirrors Python's
+// chat_model.py default of enable_thinking=False for qwen3.
+func applyQwen3ThinkingDefault(modelName string, reqBody map[string]interface{}) {
+	if !strings.Contains(strings.ToLower(modelName), "qwen3") {
+		return
+	}
+	if _, alreadySet := reqBody["enable_thinking"]; alreadySet {
+		return
+	}
+	reqBody["enable_thinking"] = false
+}
+
 // aliyunToolChoice prevents qwen-flash from repeatedly issuing another tool call
 // after a tool result has already been supplied. With "auto", qwen-flash can
 // keep emitting tool_calls even for a successful result until the ReAct graph
diff --git a/web/src/hooks/use-document-request.ts b/web/src/hooks/use-document-request.ts
index 63af8acd37..c21fea03e6 100644
--- a/web/src/hooks/use-document-request.ts
+++ b/web/src/hooks/use-document-request.ts
@@ -538,7 +538,6 @@ export const useSetDocumentPipelineParser = () => {
         parser_id: parserId,
         pipeline_id: pipelineId,
       };
-     
 
       if (parserConfig) {
         updateData.parser_config = isPipelineParserConfig(parserConfig)