diff --git a/src/api/providers/utils/__tests__/kimi-tool-call-extractor.spec.ts b/src/api/providers/utils/__tests__/kimi-tool-call-extractor.spec.ts
index d85b6cb519..85ef680a19 100644
--- a/src/api/providers/utils/__tests__/kimi-tool-call-extractor.spec.ts
+++ b/src/api/providers/utils/__tests__/kimi-tool-call-extractor.spec.ts
@@ -7,11 +7,16 @@ import {
 
 describe("kimi-tool-call-extractor", () => {
 	describe("hasKimiEmbeddedToolCalls", () => {
-		it("should return true when content contains tool call markers", () => {
+		it("should return true when content contains tool call markers with delimiters", () => {
 			const content = "Some reasoning <|tool_calls_section_begin|> stuff <|tool_calls_section_end|>"
 			expect(hasKimiEmbeddedToolCalls(content)).toBe(true)
 		})
 
+		it("should return true when content contains tool call markers without delimiters", () => {
+			const content = "Some reasoning tool_calls_section_begin stuff tool_calls_section_end"
+			expect(hasKimiEmbeddedToolCalls(content)).toBe(true)
+		})
+
 		it("should return false when content does not contain tool call markers", () => {
 			const content = "Just regular reasoning content without any tool calls"
 			expect(hasKimiEmbeddedToolCalls(content)).toBe(false)
@@ -23,7 +28,7 @@ describe("kimi-tool-call-extractor", () => {
 	})
 
 	describe("extractKimiToolCalls", () => {
-		it("should extract single tool call from reasoning content", () => {
+		it("should extract single tool call from reasoning content with delimiters", () => {
 			const content = `Some reasoning here
 <|tool_calls_section_begin|>
 <|tool_call_begin|>functions.read_file:0<|tool_call_argument_begin|>{"files":[{"path":"test.txt"}]}<|tool_call_end|>
@@ -41,7 +46,30 @@ More content after`
 					arguments: '{"files":[{"path":"test.txt"}]}',
 				},
 			})
-			expect(result.cleanedReasoningContent).not.toContain("<|tool_calls_section_begin|>")
+			expect(result.cleanedReasoningContent).not.toContain("tool_calls_section_begin")
+			expect(result.cleanedReasoningContent).toContain("Some reasoning here")
+			expect(result.cleanedReasoningContent).toContain("More content after")
+		})
+
+		it("should extract single tool call from reasoning content without delimiters", () => {
+			const content = `Some reasoning here
+tool_calls_section_begin
+tool_call_begin functions.read_file:0 tool_call_argument_begin {"files":[{"path":"test.txt"}]} tool_call_end
+tool_calls_section_end
+More content after`
+
+			const result = extractKimiToolCalls(content)
+
+			expect(result.toolCalls).toHaveLength(1)
+			expect(result.toolCalls[0]).toEqual({
+				id: "kimi-functions.read_file:0",
+				type: "function",
+				function: {
+					name: "read_file",
+					arguments: '{"files":[{"path":"test.txt"}]}',
+				},
+			})
+			expect(result.cleanedReasoningContent).not.toContain("tool_calls_section_begin")
 			expect(result.cleanedReasoningContent).toContain("Some reasoning here")
 			expect(result.cleanedReasoningContent).toContain("More content after")
 		})
@@ -110,9 +138,9 @@ More content after`
 			const result = extractKimiToolCalls(content)
 
 			// The cleaned content should not contain tool call markers
-			expect(result.cleanedReasoningContent).not.toContain("<|tool_calls_section_begin|>")
-			expect(result.cleanedReasoningContent).not.toContain("<|tool_calls_section_end|>")
-			expect(result.cleanedReasoningContent).not.toContain("<|tool_call_begin|>")
+			expect(result.cleanedReasoningContent).not.toContain("tool_calls_section_begin")
+			expect(result.cleanedReasoningContent).not.toContain("tool_calls_section_end")
+			expect(result.cleanedReasoningContent).not.toContain("tool_call_begin")
 			expect(result.cleanedReasoningContent).toContain("Before tool calls")
 			expect(result.cleanedReasoningContent).toContain("After tool calls")
 		})
@@ -147,6 +175,18 @@ End reasoning`
 			expect(result.toolCalls[0].function.name).toBe("tool1")
 			expect(result.toolCalls[1].function.name).toBe("tool2")
 		})
+
+		it("should handle markers without delimiters", () => {
+			const content = `Some reasoning
+tool_calls_section_begin
+tool_call_begin functions.read_file:0 tool_call_argument_begin {"path":"test.txt"} tool_call_end
+tool_calls_section_end`
+
+			const result = extractKimiToolCalls(content)
+
+			expect(result.toolCalls).toHaveLength(1)
+			expect(result.toolCalls[0].function.name).toBe("read_file")
+		})
 	})
 
 	describe("isKimiThinkingModel", () => {
diff --git a/src/api/providers/utils/kimi-tool-call-extractor.ts b/src/api/providers/utils/kimi-tool-call-extractor.ts
index d89546e082..740e59cf6e 100644
--- a/src/api/providers/utils/kimi-tool-call-extractor.ts
+++ b/src/api/providers/utils/kimi-tool-call-extractor.ts
@@ -6,6 +6,9 @@
  * - <|tool_call_begin|> ... <|tool_call_end|> wraps each individual tool call
  * - <|tool_call_argument_begin|> marks the start of arguments JSON
  *
+ * Note: The model may output these markers with or without the <| |> delimiters,
+ * so patterns are designed to handle both formats.
+ *
  * Format example:
  * <|tool_calls_section_begin|>
  * <|tool_call_begin|>functions.read_file:0<|tool_call_argument_begin|>{"files":[{"path":"test.txt"}]}<|tool_call_end|>
@@ -36,9 +39,11 @@ export interface KimiToolCallExtractionResult {
 
 /**
  * Checks if the content contains Kimi K2 Thinking model's embedded tool call markers.
+ * Note: The model may output markers with or without <| |> delimiters, so we check
+ * for the core marker text only.
  */
 export function hasKimiEmbeddedToolCalls(content: string): boolean {
-	return content.includes("<|tool_calls_section_begin|>")
+	return content.includes("tool_calls_section_begin")
 }
 
 /**
@@ -58,7 +63,8 @@ export function extractKimiToolCalls(content: string): KimiToolCallExtractionRes
 	const toolCalls: KimiToolCall[] = []
 
 	// Pattern to match tool call sections
-	const sectionPattern = /<\|tool_calls_section_begin\|>(.*?)<\|tool_calls_section_end\|>/gs
+	// Note: Handles both <|marker|> and marker formats (delimiters are optional)
+	const sectionPattern = /(?:<\|)?tool_calls_section_begin(?:\|>)?(.*?)(?:<\|)?tool_calls_section_end(?:\|>)?/gs
 	const toolCallSections = content.match(sectionPattern)
 
 	if (!toolCallSections || toolCallSections.length === 0) {
@@ -70,8 +76,9 @@ export function extractKimiToolCalls(content: string): KimiToolCallExtractionRes
 
 	// Pattern to extract individual tool calls
 	// Format: <|tool_call_begin|>functions.tool_name:index<|tool_call_argument_begin|>JSON_ARGS<|tool_call_end|>
+	// Note: Handles both <|marker|> and marker formats (delimiters are optional)
 	const funcCallPattern =
-		/<\|tool_call_begin\|>\s*([\w.]+:\d+)\s*<\|tool_call_argument_begin\|>\s*(.*?)\s*<\|tool_call_end\|>/gs
+		/(?:<\|)?tool_call_begin(?:\|>)?\s*([\w.]+:\d+)\s*(?:<\|)?tool_call_argument_begin(?:\|>)?\s*(.*?)\s*(?:<\|)?tool_call_end(?:\|>)?/gs
 
 	for (const section of toolCallSections) {
 		let match