diff --git a/conf/models/huaweicloud.json b/conf/models/huaweicloud.json
index 06f47636b5..b6502f14eb 100755
--- a/conf/models/huaweicloud.json
+++ b/conf/models/huaweicloud.json
@@ -20,6 +20,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -31,6 +34,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -42,6 +48,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": false,
         "clear_thinking": true
@@ -53,6 +62,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": false,
         "clear_thinking": true
@@ -63,7 +75,10 @@
       "max_tokens": 131072,
       "model_types": [
         "chat"
-      ]
+      ],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "deepseek-r1-250528",
@@ -71,6 +86,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -82,6 +100,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -93,6 +114,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -104,6 +128,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -115,6 +142,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -125,7 +155,10 @@
       "max_tokens": 131072,
       "model_types": [
         "chat"
-      ]
+      ],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "glm-5",
@@ -133,6 +166,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -144,6 +180,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -155,7 +194,10 @@
       "model_types": [
         "chat",
         "vision"
-      ]
+      ],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "bge-m3",
diff --git a/conf/models/qiniu.json b/conf/models/qiniu.json
index 51ab82f2b3..ab4f2b485b 100644
--- a/conf/models/qiniu.json
+++ b/conf/models/qiniu.json
@@ -13,6 +13,9 @@
       "name": "deepseek/deepseek-v4-flash",
       "max_tokens": 1048576,
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -22,6 +25,9 @@
       "name": "deepseek/deepseek-v4-pro",
       "max_tokens": 1048576,
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -42,6 +48,9 @@
     {
       "name": "z-ai/glm-5.1",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -50,6 +59,9 @@
     {
       "name": "z-ai/glm-5",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -58,6 +70,9 @@
     {
       "name": "minimax/minimax-m2.7",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -66,6 +81,9 @@
     {
       "name": "minimax/minimax-m2.5",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -74,6 +92,9 @@
     {
       "name": "minimax/minimax-m2.5-highspeed",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -81,12 +102,18 @@
     },
     {
       "name": "minimax/minimax-m2.1",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "kimi-k2-thinking",
       "max_tokens": 262144,
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -94,16 +121,25 @@
     },
     {
       "name": "meituan/longcat-flash-lite",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen3-max",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "z-ai/glm-4.6",
       "max_tokens": 204800,
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -111,11 +147,17 @@
     },
     {
       "name": "z-ai/glm-4.7",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "deepseek/deepseek-v3.2-251201",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -124,6 +166,9 @@
     {
       "name": "deepseek/deepseek-v3.2-exp-thinking",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -131,11 +176,17 @@
     },
     {
       "name": "deepseek/deepseek-v3.1-terminus",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "deepseek/deepseek-v3.1-terminus-thinking",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -143,15 +194,24 @@
     },
     {
       "name": "deepseek-v3.1",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "deepseek-v3-0324",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "deepseek-r1-0528",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -160,6 +220,9 @@
     {
       "name": "deepseek-r1",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -199,6 +262,9 @@
     {
       "name": "doubao-seed-2.0-lite",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -216,6 +282,9 @@
     {
       "name": "doubao-seed-2.0-code",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -224,6 +293,9 @@
     {
       "name": "qwen3-next-80b-a3b-thinking",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -232,6 +304,9 @@
     {
       "name": "qwen3-235b-a22b-thinking-2507",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -240,6 +315,9 @@
     {
       "name": "qwen3-max-2026-01-23",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -247,11 +325,17 @@
     },
     {
       "name": "qwen3-next-80b-a3b-instruct",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen3-max-preview",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen-2.5-vl-72b-instruct",
@@ -259,11 +343,17 @@
     },
     {
       "name": "qwen3-coder-480b-a35b-instruct",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen-turbo",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -271,11 +361,17 @@
     },
     {
       "name": "qwen3-235b-a22b-instruct-2507",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen3-32b",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -284,6 +380,9 @@
     {
       "name": "qwen3-30b-a3b",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -291,7 +390,10 @@
     },
     {
       "name": "qwen3-235b-a22b",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen-2.5-vl-7b-instruct",
@@ -303,12 +405,18 @@
     },
     {
       "name": "qwen2.5-max-2025-01-25",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "minimax-m1",
       "max_tokens": 1048576,
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -317,6 +425,9 @@
     {
       "name": "glm-4.5",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -328,11 +439,17 @@
     },
     {
       "name": "deepseek-v3",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen3-30b-a3b-thinking-2507",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -341,6 +458,9 @@
     {
       "name": "glm-4.5-air",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -361,6 +481,9 @@
     {
       "name": "qwen/qwen3.6-plus",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -368,11 +491,17 @@
     },
     {
       "name": "deepseek/deepseek-v3.2-exp",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen/qwen3.7-max",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -390,6 +519,9 @@
     {
       "name": "tencent/hy3-preview",
       "model_types": ["chat"],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -413,7 +545,10 @@
     },
     {
       "name": "qwen3-30b-a3b-instruct-2507",
-      "model_types": ["chat"]
+      "model_types": ["chat"],
+      "tools": {
+        "support": true
+      }
     }
   ]
 }
\ No newline at end of file
diff --git a/conf/models/siliconflow.json b/conf/models/siliconflow.json
index f1d6b523b0..c6e368dcb4 100644
--- a/conf/models/siliconflow.json
+++ b/conf/models/siliconflow.json
@@ -20,6 +20,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -31,6 +34,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -43,6 +49,9 @@
         "chat",
         "vision"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -54,6 +63,9 @@
       "model_types": [
         "chat"
       ],
+      "tools": {
+        "support": true
+      },
       "thinking": {
         "default_value": true,
         "clear_thinking": true
@@ -64,21 +76,30 @@
       "max_tokens": 32768,
       "model_types": [
         "chat"
-      ]
+      ],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "qwen/qwen3.5-4b",
       "max_tokens": 262144,
       "model_types": [
         "chat"
-      ]
+      ],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "tencent/hunyuan-mt-7b",
       "max_tokens": 32768,
       "model_types": [
         "chat"
-      ]
+      ],
+      "tools": {
+        "support": true
+      }
     },
     {
       "name": "BAAI/bge-reranker-v2-m3",
diff --git a/internal/entity/models/aliyun.go b/internal/entity/models/aliyun.go
index 74822ef71d..0d4acdcbfd 100644
--- a/internal/entity/models/aliyun.go
+++ b/internal/entity/models/aliyun.go
@@ -23,7 +23,6 @@ import (
 	"fmt"
 	"io"
 	"net/http"
-	"sort"
 	"strings"
 
 	"ragflow/internal/common"
@@ -70,12 +69,10 @@ func (a *AliyunModel) ChatWithMessages(modelName string, messages []Message, api
 
 	url := fmt.Sprintf("%s/%s", strings.TrimSuffix(baseURL, "/"), a.baseModel.URLSuffix.Chat)
 
-	apiMessages := aliyunChatMessages(messages)
-
 	// Build request body
 	reqBody := map[string]interface{}{
 		"model":       modelName,
-		"messages":    apiMessages,
+		"messages":    buildChatMessages(messages),
 		"stream":      false,
 		"temperature": 1,
 	}
@@ -168,7 +165,7 @@ func (a *AliyunModel) ChatWithMessages(modelName string, messages []Message, api
 	}
 
 	answer, hasAnswer := messageMap["content"].(string)
-	toolCalls := aliyunToolCalls(messageMap)
+	toolCalls := extractToolCalls(messageMap)
 	if !hasAnswer && len(toolCalls) == 0 {
 		return nil, fmt.Errorf("response contains neither content nor tool calls")
 	}
@@ -212,12 +209,10 @@ func (a *AliyunModel) ChatStreamlyWithSender(modelName string, messages []Messag
 
 	url := fmt.Sprintf("%s/%s", strings.TrimSuffix(baseURL, "/"), a.baseModel.URLSuffix.Chat)
 
-	apiMessages := aliyunChatMessages(messages)
-
 	// Build request body with streaming enabled
 	reqBody := map[string]interface{}{
 		"model":       modelName,
-		"messages":    apiMessages,
+		"messages":    buildChatMessages(messages),
 		"stream":      true,
 		"temperature": 1,
 	}
@@ -308,14 +303,7 @@ func (a *AliyunModel) ChatStreamlyWithSender(modelName string, messages []Messag
 			return nil
 		}
 
-		if toolCalls, ok := delta["tool_calls"].([]interface{}); ok {
-			for _, rawToolCall := range toolCalls {
-				toolCall, ok := rawToolCall.(map[string]interface{})
-				if !ok {
-					continue
-				}
-				mergeAliyunToolCallDelta(accumulatedToolCalls, toolCall)
-			}
+		if accumulateToolCallDeltas(delta, accumulatedToolCalls) {
 			return nil
 		}
 
@@ -342,78 +330,13 @@ func (a *AliyunModel) ChatStreamlyWithSender(modelName string, messages []Messag
 		return fmt.Errorf("aliyun: stream ended before [DONE] or finish_reason")
 	}
 
-	if len(accumulatedToolCalls) > 0 && chatModelConfig != nil {
-		indices := make([]int, 0, len(accumulatedToolCalls))
-		for index := range accumulatedToolCalls {
-			indices = append(indices, index)
-		}
-		sort.Ints(indices)
-		toolCalls := make([]map[string]interface{}, 0, len(indices))
-		for _, index := range indices {
-			toolCalls = append(toolCalls, accumulatedToolCalls[index])
-		}
-		chatModelConfig.ToolCallsResult = &toolCalls
-	}
+	setSortedToolCallsResult(chatModelConfig, accumulatedToolCalls)
 
 	// Send [DONE] marker for OpenAI compatibility
 	endOfStream := "[DONE]"
 	return sender(&endOfStream, nil)
 }
 
-func mergeAliyunToolCallDelta(accumulated map[int]map[string]interface{}, delta map[string]interface{}) {
-	indexValue, ok := delta["index"].(float64)
-	if !ok {
-		return
-	}
-	index := int(indexValue)
-	toolCall, exists := accumulated[index]
-	if !exists {
-		toolCall = map[string]interface{}{"index": indexValue}
-		accumulated[index] = toolCall
-	}
-
-	for _, field := range []string{"id", "type"} {
-		if value, ok := delta[field].(string); ok && value != "" {
-			toolCall[field] = value
-		}
-	}
-
-	functionDelta, ok := delta["function"].(map[string]interface{})
-	if !ok {
-		return
-	}
-	function, ok := toolCall["function"].(map[string]interface{})
-	if !ok {
-		function = make(map[string]interface{})
-		toolCall["function"] = function
-	}
-	if name, ok := functionDelta["name"].(string); ok && name != "" {
-		function["name"] = name
-	}
-	if arguments, ok := functionDelta["arguments"].(string); ok {
-		currentArguments, _ := function["arguments"].(string)
-		function["arguments"] = currentArguments + arguments
-	}
-}
-
-func aliyunChatMessages(messages []Message) []map[string]interface{} {
-	apiMessages := make([]map[string]interface{}, len(messages))
-	for i, msg := range messages {
-		apiMessage := map[string]interface{}{
-			"role":    msg.Role,
-			"content": msg.Content,
-		}
-		if msg.ToolCallID != "" {
-			apiMessage["tool_call_id"] = msg.ToolCallID
-		}
-		if len(msg.ToolCalls) > 0 {
-			apiMessage["tool_calls"] = msg.ToolCalls
-		}
-		apiMessages[i] = apiMessage
-	}
-	return apiMessages
-}
-
 // 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
@@ -438,20 +361,6 @@ func aliyunToolChoice(modelName string, messages []Message, configured *string)
 	return choice
 }
 
-func aliyunToolCalls(message map[string]interface{}) []map[string]interface{} {
-	rawToolCalls, ok := message["tool_calls"].([]interface{})
-	if !ok {
-		return nil
-	}
-	toolCalls := make([]map[string]interface{}, 0, len(rawToolCalls))
-	for _, rawToolCall := range rawToolCalls {
-		if toolCall, ok := rawToolCall.(map[string]interface{}); ok {
-			toolCalls = append(toolCalls, toolCall)
-		}
-	}
-	return toolCalls
-}
-
 type aliyunEmbeddingResponse struct {
 	Data   []EmbeddingData `json:"data"`
 	Model  string          `json:"model"`
diff --git a/internal/entity/models/base_model.go b/internal/entity/models/base_model.go
index f820f6cf88..3775a9579c 100644
--- a/internal/entity/models/base_model.go
+++ b/internal/entity/models/base_model.go
@@ -240,6 +240,32 @@ func buildChatMessages(messages []Message) []map[string]any {
 	return apiMessages
 }
 
+// applyChatToolConfig adds OpenAI-compatible tool configuration to a request.
+func applyChatToolConfig(reqBody map[string]interface{}, chatConfig *ChatConfig) {
+	if chatConfig == nil || chatConfig.Tools == nil {
+		return
+	}
+	reqBody["tools"] = chatConfig.Tools
+	if chatConfig.ToolChoice != nil {
+		reqBody["tool_choice"] = *chatConfig.ToolChoice
+	}
+}
+
+// extractToolCalls converts an OpenAI-compatible message's tool calls.
+func extractToolCalls(message map[string]interface{}) []map[string]interface{} {
+	rawToolCalls, ok := message["tool_calls"].([]interface{})
+	if !ok {
+		return nil
+	}
+	toolCalls := make([]map[string]interface{}, 0, len(rawToolCalls))
+	for _, rawToolCall := range rawToolCalls {
+		if toolCall, ok := rawToolCall.(map[string]interface{}); ok {
+			toolCalls = append(toolCalls, toolCall)
+		}
+	}
+	return toolCalls
+}
+
 // setSortedToolCallsResult stores accumulated tool calls in index order.
 func setSortedToolCallsResult(chatConfig *ChatConfig, accumulatedToolCalls map[int]map[string]any) {
 	if chatConfig == nil || len(accumulatedToolCalls) == 0 {
diff --git a/internal/entity/models/huaweicloud.go b/internal/entity/models/huaweicloud.go
index a4795ad714..8cf75c9eb2 100644
--- a/internal/entity/models/huaweicloud.go
+++ b/internal/entity/models/huaweicloud.go
@@ -63,17 +63,6 @@ func huaweiCloudRegionForModel(api *APIConfig, modelName string) string {
 	return region
 }
 
-func huaweiCloudMessages(messages []Message) []map[string]interface{} {
-	apiMessages := make([]map[string]interface{}, len(messages))
-	for i, msg := range messages {
-		apiMessages[i] = map[string]interface{}{
-			"role":    msg.Role,
-			"content": msg.Content,
-		}
-	}
-	return apiMessages
-}
-
 func huaweiCloudIsDeepSeekV4(modelName string) bool {
 	model := strings.ToLower(modelName)
 	return strings.Contains(model, "deepseek-v4-pro") || strings.Contains(model, "deepseek-v4-flash")
@@ -200,10 +189,11 @@ func (h *HuaweiCloudModel) ChatWithMessages(modelName string, messages []Message
 
 	reqb := map[string]interface{}{
 		"model":    huaweiCloudChatModelName(modelName),
-		"messages": huaweiCloudMessages(messages),
+		"messages": buildChatMessages(messages),
 		"stream":   false,
 	}
 	huaweiCloudApplyChatConfig(reqb, modelName, chatModelConfig)
+	applyChatToolConfig(reqb, chatModelConfig)
 
 	jsonData, err := json.Marshal(reqb)
 	if err != nil {
@@ -253,8 +243,9 @@ func (h *HuaweiCloudModel) ChatWithMessages(modelName string, messages []Message
 	if !ok {
 		return nil, fmt.Errorf("invalid message format")
 	}
-	content, ok := messageMap["content"].(string)
-	if !ok {
+	toolCalls := extractToolCalls(messageMap)
+	content, hasContent := messageMap["content"].(string)
+	if !hasContent && len(toolCalls) == 0 {
 		return nil, fmt.Errorf("invalid content format")
 	}
 
@@ -269,6 +260,7 @@ func (h *HuaweiCloudModel) ChatWithMessages(modelName string, messages []Message
 	return &ChatResponse{
 		Answer:        &content,
 		ReasonContent: &reasonContent,
+		ToolCalls:     toolCalls,
 	}, nil
 }
 
@@ -301,13 +293,17 @@ func (h *HuaweiCloudModel) ChatStreamlyWithSender(modelName string, messages []M
 
 	reqBody := map[string]interface{}{
 		"model":    huaweiCloudChatModelName(modelName),
-		"messages": huaweiCloudMessages(messages),
+		"messages": buildChatMessages(messages),
 		"stream":   true,
 	}
 	if chatModelConfig != nil && chatModelConfig.Stream != nil && !*chatModelConfig.Stream {
 		return fmt.Errorf("stream must be true in ChatStreamlyWithSender")
 	}
 	huaweiCloudApplyChatConfig(reqBody, modelName, chatModelConfig)
+	if chatModelConfig != nil {
+		chatModelConfig.ToolCallsResult = nil
+	}
+	applyChatToolConfig(reqBody, chatModelConfig)
 
 	jsonData, err := json.Marshal(reqBody)
 	if err != nil {
@@ -336,6 +332,7 @@ func (h *HuaweiCloudModel) ChatStreamlyWithSender(modelName string, messages []M
 	}
 
 	sawTerminal := false
+	accumulatedToolCalls := make(map[int]map[string]interface{})
 	done, err := ParseSSEStream[map[string]interface{}](resp.Body, func(event map[string]interface{}) error {
 		if apiErr, ok := event["error"]; ok {
 			return fmt.Errorf("huaweicloud: upstream stream error: %v", apiErr)
@@ -349,10 +346,16 @@ func (h *HuaweiCloudModel) ChatStreamlyWithSender(modelName string, messages []M
 		if !ok {
 			return nil
 		}
+		if finishReason, ok := firstChoice["finish_reason"].(string); ok && finishReason != "" {
+			sawTerminal = true
+		}
 		delta, ok := firstChoice["delta"].(map[string]interface{})
 		if !ok {
 			return nil
 		}
+		if accumulateToolCallDeltas(delta, accumulatedToolCalls) {
+			return nil
+		}
 
 		if r, ok := delta["reasoning_content"].(string); ok && r != "" {
 			if err := sender(nil, &r); err != nil {
@@ -364,9 +367,6 @@ func (h *HuaweiCloudModel) ChatStreamlyWithSender(modelName string, messages []M
 				return err
 			}
 		}
-		if finishReason, ok := firstChoice["finish_reason"].(string); ok && finishReason != "" {
-			sawTerminal = true
-		}
 		return nil
 	})
 	if err != nil {
@@ -375,6 +375,7 @@ func (h *HuaweiCloudModel) ChatStreamlyWithSender(modelName string, messages []M
 	if !done && !sawTerminal {
 		return fmt.Errorf("huaweicloud: stream ended before [DONE] or finish_reason")
 	}
+	setSortedToolCallsResult(chatModelConfig, accumulatedToolCalls)
 
 	endOfStream := "[DONE]"
 	if err := sender(&endOfStream, nil); err != nil {
diff --git a/internal/entity/models/huaweicloud_test.go b/internal/entity/models/huaweicloud_test.go
new file mode 100644
index 0000000000..8c4d59cfc8
--- /dev/null
+++ b/internal/entity/models/huaweicloud_test.go
@@ -0,0 +1,31 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+package models
+
+import "testing"
+
+func TestHuaweiCloudToolCalls(t *testing.T) {
+	newDriver := func(baseURL string) ModelDriver {
+		return NewHuaweiCloudModel(map[string]string{"default": baseURL}, URLSuffix{Chat: "v2/chat/completions"})
+	}
+	t.Run("non-streaming", func(t *testing.T) {
+		testNonStreamingToolCall(t, "deepseek-v4-pro", "/v2/chat/completions", newDriver)
+	})
+	t.Run("streaming", func(t *testing.T) {
+		testStreamingToolCall(t, "deepseek-v4-pro", "/v2/chat/completions", newDriver)
+	})
+}
diff --git a/internal/entity/models/jina.go b/internal/entity/models/jina.go
index b10e803915..d68a0bf595 100644
--- a/internal/entity/models/jina.go
+++ b/internal/entity/models/jina.go
@@ -73,17 +73,9 @@ func (j *JinaModel) ChatWithMessages(modelName string, messages []Message, apiCo
 	baseURL = strings.TrimSuffix(baseURL, "/")
 	url := fmt.Sprintf("%s/%s", baseURL, j.baseModel.URLSuffix.Chat)
 
-	apiMessages := make([]map[string]interface{}, len(messages))
-	for i, msg := range messages {
-		apiMessages[i] = map[string]interface{}{
-			"role":    msg.Role,
-			"content": msg.Content,
-		}
-	}
-
 	reqBody := map[string]interface{}{
 		"model":    modelName,
-		"messages": apiMessages,
+		"messages": buildChatMessages(messages),
 		"stream":   false,
 	}
 
@@ -100,6 +92,7 @@ func (j *JinaModel) ChatWithMessages(modelName string, messages []Message, apiCo
 		if chatModelConfig.Stop != nil {
 			reqBody["stop"] = *chatModelConfig.Stop
 		}
+		applyChatToolConfig(reqBody, chatModelConfig)
 	}
 
 	jsonData, err := json.Marshal(reqBody)
@@ -153,8 +146,9 @@ func (j *JinaModel) ChatWithMessages(modelName string, messages []Message, apiCo
 		return nil, fmt.Errorf("invalid message format")
 	}
 
-	content, ok := messageMap["content"].(string)
-	if !ok {
+	toolCalls := extractToolCalls(messageMap)
+	content, hasContent := messageMap["content"].(string)
+	if !hasContent && len(toolCalls) == 0 {
 		return nil, fmt.Errorf("invalid content format")
 	}
 
@@ -162,6 +156,7 @@ func (j *JinaModel) ChatWithMessages(modelName string, messages []Message, apiCo
 	return &ChatResponse{
 		Answer:        &content,
 		ReasonContent: &reasonContent,
+		ToolCalls:     toolCalls,
 	}, nil
 }
 
diff --git a/internal/entity/models/jina_test.go b/internal/entity/models/jina_test.go
index 1f40dc1918..26389771d8 100644
--- a/internal/entity/models/jina_test.go
+++ b/internal/entity/models/jina_test.go
@@ -89,6 +89,12 @@ func TestJinaChatHappyPath(t *testing.T) {
 	}
 }
 
+func TestJinaChatSupportsToolCalls(t *testing.T) {
+	testNonStreamingToolCall(t, "jina-vlm", "/chat/completions", func(baseURL string) ModelDriver {
+		return newJinaForTest(baseURL)
+	})
+}
+
 func TestJinaChatPropagatesConfig(t *testing.T) {
 	srv := newJinaServer(t, "/chat/completions", func(t *testing.T, body map[string]interface{}, w http.ResponseWriter) {
 		if body["max_tokens"] != float64(128) {
diff --git a/internal/entity/models/qiniu.go b/internal/entity/models/qiniu.go
index 219fabd4cb..6f6e16cfe4 100644
--- a/internal/entity/models/qiniu.go
+++ b/internal/entity/models/qiniu.go
@@ -154,16 +154,9 @@ func (q *QiniuModel) ChatWithMessages(modelName string, messages []Message, apiC
 	}
 	url := fmt.Sprintf("%s/%s", resolvedBaseURL, q.baseModel.URLSuffix.Chat)
 
-	apiMessages := make([]map[string]interface{}, len(messages))
-	for i, msg := range messages {
-		apiMessages[i] = map[string]interface{}{
-			"role":    msg.Role,
-			"content": msg.Content,
-		}
-	}
 	reqBody := map[string]interface{}{
 		"model":       modelName,
-		"messages":    apiMessages,
+		"messages":    buildChatMessages(messages),
 		"stream":      false,
 		"temperature": 1,
 	}
@@ -186,6 +179,7 @@ func (q *QiniuModel) ChatWithMessages(modelName string, messages []Message, apiC
 		}
 
 		applyQiniuThinkingConfig(reqBody, modelName, chatModelConfig)
+		applyChatToolConfig(reqBody, chatModelConfig)
 	}
 
 	jsonData, err := json.Marshal(reqBody)
@@ -239,15 +233,16 @@ func (q *QiniuModel) ChatWithMessages(modelName string, messages []Message, apiC
 		return nil, fmt.Errorf("invalid message format")
 	}
 
-	content, ok := messageMap["content"].(string)
-	if !ok {
+	toolCalls := extractToolCalls(messageMap)
+	content, hasContent := messageMap["content"].(string)
+	if !hasContent && len(toolCalls) == 0 {
 		return nil, fmt.Errorf("invalid content format")
 	}
 
 	var reasonContent string
 	if chatModelConfig != nil && chatModelConfig.Thinking != nil && *chatModelConfig.Thinking {
 		reasonContent, ok = messageMap["reasoning_content"].(string)
-		if !ok {
+		if !ok && len(toolCalls) == 0 {
 			return nil, fmt.Errorf("invalid reasoning content format")
 		}
 
@@ -259,6 +254,7 @@ func (q *QiniuModel) ChatWithMessages(modelName string, messages []Message, apiC
 	chatResponse := &ChatResponse{
 		Answer:        &content,
 		ReasonContent: &reasonContent,
+		ToolCalls:     toolCalls,
 	}
 
 	return chatResponse, nil
@@ -279,16 +275,9 @@ func (q *QiniuModel) ChatStreamlyWithSender(modelName string, messages []Message
 	baseURL := strings.TrimSuffix(resolvedBaseURL, "/")
 	url := fmt.Sprintf("%s/%s", baseURL, q.baseModel.URLSuffix.Chat)
 
-	apiMessages := make([]map[string]interface{}, len(messages))
-	for i, msg := range messages {
-		apiMessages[i] = map[string]interface{}{
-			"role":    msg.Role,
-			"content": msg.Content,
-		}
-	}
 	reqBody := map[string]interface{}{
 		"model":       modelName,
-		"messages":    apiMessages,
+		"messages":    buildChatMessages(messages),
 		"stream":      true,
 		"temperature": 1,
 	}
@@ -311,6 +300,8 @@ func (q *QiniuModel) ChatStreamlyWithSender(modelName string, messages []Message
 		}
 
 		applyQiniuThinkingConfig(reqBody, modelName, chatModelConfig)
+		chatModelConfig.ToolCallsResult = nil
+		applyChatToolConfig(reqBody, chatModelConfig)
 	}
 	jsonData, err := json.Marshal(reqBody)
 	if err != nil {
@@ -339,7 +330,9 @@ func (q *QiniuModel) ChatStreamlyWithSender(modelName string, messages []Message
 		return fmt.Errorf("API request failed with status %d: %s", resp.StatusCode, string(body))
 	}
 
-	if _, err := ParseSSEStream[map[string]interface{}](resp.Body, func(event map[string]interface{}) error {
+	sawTerminal := false
+	accumulatedToolCalls := make(map[int]map[string]interface{})
+	done, err := ParseSSEStream[map[string]interface{}](resp.Body, func(event map[string]interface{}) error {
 		common.Info(fmt.Sprintf("%v", event))
 
 		choices, ok := event["choices"].([]interface{})
@@ -350,10 +343,16 @@ func (q *QiniuModel) ChatStreamlyWithSender(modelName string, messages []Message
 		if !ok {
 			return nil
 		}
+		if finishReason, ok := firstChoice["finish_reason"].(string); ok && finishReason != "" {
+			sawTerminal = true
+		}
 		delta, ok := firstChoice["delta"].(map[string]interface{})
 		if !ok {
 			return nil
 		}
+		if accumulateToolCallDeltas(delta, accumulatedToolCalls) {
+			return nil
+		}
 
 		reasoningContent, ok := delta["reasoning_content"].(string)
 		if ok && reasoningContent != "" {
@@ -369,9 +368,14 @@ func (q *QiniuModel) ChatStreamlyWithSender(modelName string, messages []Message
 		}
 
 		return nil
-	}); err != nil {
+	})
+	if err != nil {
 		return fmt.Errorf("failed to scan response body: %w", err)
 	}
+	if !done && !sawTerminal {
+		return fmt.Errorf("qiniu: stream ended before [DONE] or finish_reason")
+	}
+	setSortedToolCallsResult(chatModelConfig, accumulatedToolCalls)
 	// Send [DONE] marker for OpenAI compatibility
 	endOfStream := "[DONE]"
 	if err = sender(&endOfStream, nil); err != nil {
@@ -465,7 +469,11 @@ func (q *QiniuModel) Balance(apiConfig *APIConfig) (map[string]interface{}, erro
 }
 
 func (q *QiniuModel) CheckConnection(apiConfig *APIConfig) error {
-	return fmt.Errorf("%s, no such method", q.Name())
+	_, err := q.ListModels(apiConfig)
+	if err != nil {
+		return err
+	}
+	return nil
 }
 
 func (q *QiniuModel) ListTasks(apiConfig *APIConfig) ([]ListTaskStatus, error) {
diff --git a/internal/entity/models/qiniu_test.go b/internal/entity/models/qiniu_test.go
new file mode 100644
index 0000000000..a0436a927f
--- /dev/null
+++ b/internal/entity/models/qiniu_test.go
@@ -0,0 +1,87 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+package models
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+)
+
+func TestQiniuToolCalls(t *testing.T) {
+	newDriver := func(baseURL string) ModelDriver {
+		return NewQiniuModel(map[string]string{"default": baseURL}, URLSuffix{Chat: "chat/completions"})
+	}
+	t.Run("non-streaming", func(t *testing.T) {
+		testNonStreamingToolCall(t, "deepseek/deepseek-v4-pro", "/chat/completions", newDriver)
+	})
+	t.Run("streaming", func(t *testing.T) {
+		testStreamingToolCall(t, "deepseek/deepseek-v4-pro", "/chat/completions", newDriver)
+	})
+}
+
+func TestQiniuChatStreamRejectsTruncatedResponse(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		_, _ = w.Write([]byte(`data: {"choices":[{"delta":{"content":"partial"}}]}` + "\n\n"))
+	}))
+	defer server.Close()
+
+	apiKey := "test-key"
+	model := NewQiniuModel(map[string]string{"default": server.URL}, URLSuffix{Chat: "chat/completions"})
+	err := model.ChatStreamlyWithSender("model", []Message{{Role: "user", Content: "hi"}}, &APIConfig{ApiKey: &apiKey}, &ChatConfig{}, nil, func(_, _ *string) error { return nil })
+	if err == nil || !strings.Contains(err.Error(), "stream ended before [DONE] or finish_reason") {
+		t.Fatalf("error = %v", err)
+	}
+}
+
+func TestQiniuCheckConnectionUsesListModels(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodGet {
+			t.Errorf("method = %s, want GET", r.Method)
+		}
+		if r.URL.Path != "/models" {
+			t.Errorf("path = %s, want /models", r.URL.Path)
+		}
+		if got := r.Header.Get("Authorization"); got != "Bearer test-key" {
+			t.Errorf("Authorization = %q, want %q", got, "Bearer test-key")
+		}
+		_, _ = w.Write([]byte(`{"object":"list","data":[{"id":"deepseek/deepseek-v4-flash"}]}`))
+	}))
+	defer server.Close()
+
+	apiKey := "test-key"
+	model := NewQiniuModel(map[string]string{"default": server.URL}, URLSuffix{Models: "models"})
+	if err := model.CheckConnection(&APIConfig{ApiKey: &apiKey}); err != nil {
+		t.Fatalf("CheckConnection() error = %v", err)
+	}
+}
+
+func TestQiniuCheckConnectionPropagatesListModelsError(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		http.Error(w, `{"message":"invalid api key"}`, http.StatusUnauthorized)
+	}))
+	defer server.Close()
+
+	apiKey := "invalid-key"
+	model := NewQiniuModel(map[string]string{"default": server.URL}, URLSuffix{Models: "models"})
+	err := model.CheckConnection(&APIConfig{ApiKey: &apiKey})
+	if err == nil || !strings.Contains(err.Error(), "status 401") {
+		t.Fatalf("CheckConnection() error = %v, want status 401", err)
+	}
+}
diff --git a/internal/entity/models/siliconflow.go b/internal/entity/models/siliconflow.go
index 3abfbc3663..cb6196743f 100644
--- a/internal/entity/models/siliconflow.go
+++ b/internal/entity/models/siliconflow.go
@@ -82,19 +82,10 @@ func (s *SiliconflowModel) ChatWithMessages(modelName string, messages []Message
 	}
 	url := fmt.Sprintf("%s/%s", resolvedBaseURL, s.baseModel.URLSuffix.Chat)
 
-	// Convert messages to the format expected by API
-	apiMessages := make([]map[string]interface{}, len(messages))
-	for i, msg := range messages {
-		apiMessages[i] = map[string]interface{}{
-			"role":    msg.Role,
-			"content": msg.Content,
-		}
-	}
-
 	// Build request body
 	reqBody := map[string]interface{}{
 		"model":    modelName,
-		"messages": apiMessages,
+		"messages": buildChatMessages(messages),
 		"stream":   false,
 	}
 
@@ -126,6 +117,8 @@ func (s *SiliconflowModel) ChatWithMessages(modelName string, messages []Message
 			// by SiliconFlow, breaking the thinking feature).
 			reqBody["enable_thinking"] = *chatModelConfig.Thinking
 		}
+
+		applyChatToolConfig(reqBody, chatModelConfig)
 	}
 
 	// Qwen3 family: disable thinking by default (matches Python's
@@ -186,8 +179,9 @@ func (s *SiliconflowModel) ChatWithMessages(modelName string, messages []Message
 		return nil, fmt.Errorf("invalid message format")
 	}
 
-	content, ok := messageMap["content"].(string)
-	if !ok {
+	toolCalls := extractToolCalls(messageMap)
+	content, hasContent := messageMap["content"].(string)
+	if !hasContent && len(toolCalls) == 0 {
 		return nil, fmt.Errorf("invalid content format")
 	}
 
@@ -212,6 +206,7 @@ func (s *SiliconflowModel) ChatWithMessages(modelName string, messages []Message
 	chatResponse := &ChatResponse{
 		Answer:        &content,
 		ReasonContent: &reasonContent,
+		ToolCalls:     toolCalls,
 	}
 
 	return chatResponse, nil
@@ -233,19 +228,10 @@ func (s *SiliconflowModel) ChatStreamlyWithSender(modelName string, messages []M
 	}
 	url := fmt.Sprintf("%s/%s", resolvedBaseURL, s.baseModel.URLSuffix.Chat)
 
-	// Convert messages to API format
-	apiMessages := make([]map[string]interface{}, len(messages))
-	for i, msg := range messages {
-		apiMessages[i] = map[string]interface{}{
-			"role":    msg.Role,
-			"content": msg.Content,
-		}
-	}
-
 	// Build request body with streaming enabled
 	reqBody := map[string]interface{}{
 		"model":    modelName,
-		"messages": apiMessages,
+		"messages": buildChatMessages(messages),
 		"stream":   true,
 	}
 
@@ -273,6 +259,9 @@ func (s *SiliconflowModel) ChatStreamlyWithSender(modelName string, messages []M
 		if chatModelConfig.Stop != nil {
 			reqBody["stop"] = *chatModelConfig.Stop
 		}
+
+		chatModelConfig.ToolCallsResult = nil
+		applyChatToolConfig(reqBody, chatModelConfig)
 	}
 
 	// Qwen3 family: disable thinking by default (matches Python's
@@ -310,6 +299,7 @@ func (s *SiliconflowModel) ChatStreamlyWithSender(modelName string, messages []M
 
 	// SSE parsing: read line by line
 	sawTerminal := false
+	accumulatedToolCalls := make(map[int]map[string]interface{})
 	done, err := ParseSSEStream[map[string]interface{}](resp.Body, func(event map[string]interface{}) error {
 		common.Info(fmt.Sprintf("%v", event))
 
@@ -322,11 +312,17 @@ func (s *SiliconflowModel) ChatStreamlyWithSender(modelName string, messages []M
 		if !ok {
 			return nil
 		}
+		if finishReason, ok := firstChoice["finish_reason"].(string); ok && finishReason != "" {
+			sawTerminal = true
+		}
 
 		delta, ok := firstChoice["delta"].(map[string]interface{})
 		if !ok {
 			return nil
 		}
+		if accumulateToolCallDeltas(delta, accumulatedToolCalls) {
+			return nil
+		}
 
 		reasoningContent, ok := delta["reasoning_content"].(string)
 		if ok && reasoningContent != "" {
@@ -342,11 +338,6 @@ func (s *SiliconflowModel) ChatStreamlyWithSender(modelName string, messages []M
 			}
 		}
 
-		finishReason, ok := firstChoice["finish_reason"].(string)
-		if ok && finishReason != "" {
-			sawTerminal = true
-		}
-
 		return nil
 	})
 	if err != nil {
@@ -355,6 +346,7 @@ func (s *SiliconflowModel) ChatStreamlyWithSender(modelName string, messages []M
 	if !done && !sawTerminal {
 		return fmt.Errorf("siliconflow: stream ended before [DONE] or finish_reason")
 	}
+	setSortedToolCallsResult(chatModelConfig, accumulatedToolCalls)
 
 	// Send [DONE] marker for OpenAI compatibility
 	endOfStream := "[DONE]"
diff --git a/internal/entity/models/siliconflow_test.go b/internal/entity/models/siliconflow_test.go
new file mode 100644
index 0000000000..28bb872478
--- /dev/null
+++ b/internal/entity/models/siliconflow_test.go
@@ -0,0 +1,48 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+package models
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+func TestSiliconflowToolCalls(t *testing.T) {
+	newDriver := func(baseURL string) ModelDriver {
+		return NewSiliconflowModel(map[string]string{"default": baseURL}, URLSuffix{Chat: "chat/completions"})
+	}
+	t.Run("non-streaming", func(t *testing.T) {
+		testNonStreamingToolCall(t, "Pro/deepseek-ai/DeepSeek-V4-Pro", "/chat/completions", newDriver)
+	})
+	t.Run("streaming", func(t *testing.T) {
+		testStreamingToolCall(t, "Pro/deepseek-ai/DeepSeek-V4-Pro", "/chat/completions", newDriver)
+	})
+}
+
+func TestSiliconflowChatRejectsMissingContent(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		_, _ = w.Write([]byte(`{"choices":[{"message":{}}]}`))
+	}))
+	defer server.Close()
+
+	apiKey := "test-key"
+	model := NewSiliconflowModel(map[string]string{"default": server.URL}, URLSuffix{Chat: "chat/completions"})
+	if _, err := model.ChatWithMessages("model", []Message{{Role: "user", Content: "hi"}}, &APIConfig{ApiKey: &apiKey}, nil, nil); err == nil {
+		t.Fatal("expected missing content error")
+	}
+}
diff --git a/internal/entity/models/tool_call_test.go b/internal/entity/models/tool_call_test.go
new file mode 100644
index 0000000000..6bedbd911c
--- /dev/null
+++ b/internal/entity/models/tool_call_test.go
@@ -0,0 +1,106 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+package models
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+func testNonStreamingToolCall(t *testing.T, modelName, path string, newDriver func(string) ModelDriver) {
+	t.Helper()
+	var requestBody map[string]interface{}
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != path {
+			t.Errorf("path = %q, want %q", r.URL.Path, path)
+		}
+		if err := json.NewDecoder(r.Body).Decode(&requestBody); err != nil {
+			t.Errorf("decode request: %v", err)
+		}
+		_, _ = w.Write([]byte(`{"choices":[{"message":{"content":null,"tool_calls":[{"id":"call-2","type":"function","function":{"name":"retrieval","arguments":"{}"}}]}}]}`))
+	}))
+	defer server.Close()
+
+	apiKey, toolChoice := "test-key", "required"
+	response, err := newDriver(server.URL).ChatWithMessages(
+		modelName,
+		[]Message{
+			{Role: "assistant", ToolCalls: []map[string]interface{}{{"id": "call-1"}}},
+			{Role: "tool", Content: "result", ToolCallID: "call-1"},
+		},
+		&APIConfig{ApiKey: &apiKey},
+		&ChatConfig{Tools: []map[string]interface{}{{"type": "function"}}, ToolChoice: &toolChoice},
+		nil,
+	)
+	if err != nil {
+		t.Fatalf("ChatWithMessages: %v", err)
+	}
+	if requestBody["tool_choice"] != toolChoice || requestBody["tools"] == nil {
+		t.Errorf("tool config = %#v", requestBody)
+	}
+	messages, _ := requestBody["messages"].([]interface{})
+	toolMessage, _ := messages[1].(map[string]interface{})
+	if toolMessage["tool_call_id"] != "call-1" {
+		t.Errorf("tool message = %#v", toolMessage)
+	}
+	if len(response.ToolCalls) != 1 || response.ToolCalls[0]["id"] != "call-2" {
+		t.Errorf("tool calls = %#v", response.ToolCalls)
+	}
+}
+
+func testStreamingToolCall(t *testing.T, modelName, path string, newDriver func(string) ModelDriver) {
+	t.Helper()
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != path {
+			t.Errorf("path = %q, want %q", r.URL.Path, path)
+		}
+		w.Header().Set("Content-Type", "text/event-stream")
+		_, _ = w.Write([]byte(`data: {"choices":[{"delta":{"tool_calls":[{"index":0,"id":"call-1","type":"function","function":{"name":"retrieval","arguments":"{\"query\":\""}}]}}]}
+
+data: {"choices":[{"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ragflow\"}"}}]},"finish_reason":"tool_calls"}]}
+
+`))
+	}))
+	defer server.Close()
+
+	apiKey, toolChoice := "test-key", "auto"
+	config := &ChatConfig{
+		Tools:           []map[string]interface{}{{"type": "function"}},
+		ToolChoice:      &toolChoice,
+		ToolCallsResult: &[]map[string]interface{}{{"id": "stale"}},
+	}
+	err := newDriver(server.URL).ChatStreamlyWithSender(
+		modelName,
+		[]Message{{Role: "user", Content: "find ragflow"}},
+		&APIConfig{ApiKey: &apiKey},
+		config,
+		nil,
+		func(_, _ *string) error { return nil },
+	)
+	if err != nil {
+		t.Fatalf("ChatStreamlyWithSender: %v", err)
+	}
+	if config.ToolCallsResult == nil || len(*config.ToolCallsResult) != 1 {
+		t.Fatalf("tool calls = %#v", config.ToolCallsResult)
+	}
+	function, _ := (*config.ToolCallsResult)[0]["function"].(map[string]interface{})
+	if function["arguments"] != `{"query":"ragflow"}` {
+		t.Errorf("function = %#v", function)
+	}
+}