diff --git a/src/api/providers/__tests__/openai-native-tools.spec.ts b/src/api/providers/__tests__/openai-native-tools.spec.ts
index 4ce6afd320..b3c0ae0dfe 100644
--- a/src/api/providers/__tests__/openai-native-tools.spec.ts
+++ b/src/api/providers/__tests__/openai-native-tools.spec.ts
@@ -296,4 +296,83 @@ describe("OpenAiNativeHandler MCP tool schema handling", () => {
 		expect(tool.parameters.properties.metadata.additionalProperties).toBe(false) // Nested object
 		expect(tool.parameters.properties.metadata.properties.labels.items.additionalProperties).toBe(false) // Array items
 	})
+
+	it("should handle missing call_id and name in tool_call_arguments.delta by using pending tool identity", async () => {
+		const handler = new OpenAiNativeHandler({
+			openAiNativeApiKey: "test-key",
+			apiModelId: "gpt-4o",
+		} as ApiHandlerOptions)
+
+		const mockClient = {
+			responses: {
+				create: vi.fn().mockImplementation(() => {
+					return {
+						[Symbol.asyncIterator]: async function* () {
+							// 1. Emit output_item.added with tool identity
+							yield {
+								type: "response.output_item.added",
+								item: {
+									type: "function_call",
+									call_id: "call_123",
+									name: "read_file",
+									arguments: "",
+								},
+							}
+
+							// 2. Emit tool_call_arguments.delta WITHOUT identity (just args)
+							yield {
+								type: "response.function_call_arguments.delta",
+								delta: '{"path":',
+							}
+
+							// 3. Emit another delta
+							yield {
+								type: "response.function_call_arguments.delta",
+								delta: '"/tmp/test.txt"}',
+							}
+
+							// 4. Emit output_item.done
+							yield {
+								type: "response.output_item.done",
+								item: {
+									type: "function_call",
+									call_id: "call_123",
+									name: "read_file",
+									arguments: '{"path":"/tmp/test.txt"}',
+								},
+							}
+						},
+					}
+				}),
+			},
+		}
+		;(handler as any).client = mockClient
+
+		const stream = handler.createMessage("system prompt", [], {
+			taskId: "test-task-id",
+		})
+
+		const chunks: any[] = []
+		for await (const chunk of stream) {
+			if (chunk.type === "tool_call_partial") {
+				chunks.push(chunk)
+			}
+		}
+
+		expect(chunks.length).toBe(2)
+		expect(chunks[0]).toEqual({
+			type: "tool_call_partial",
+			index: 0,
+			id: "call_123", // Should be filled from pendingToolCallId
+			name: "read_file", // Should be filled from pendingToolCallName
+			arguments: '{"path":',
+		})
+		expect(chunks[1]).toEqual({
+			type: "tool_call_partial",
+			index: 0,
+			id: "call_123",
+			name: "read_file",
+			arguments: '"/tmp/test.txt"}',
+		})
+	})
 })
diff --git a/src/api/providers/openai-native.ts b/src/api/providers/openai-native.ts
index 58a62497f7..cd5c3bbcd7 100644
--- a/src/api/providers/openai-native.ts
+++ b/src/api/providers/openai-native.ts
@@ -32,6 +32,13 @@ export class OpenAiNativeHandler extends BaseProvider implements SingleCompletio
 	protected options: ApiHandlerOptions
 	private client: OpenAI
 	private readonly providerName = "OpenAI Native"
+	/**
+	 * Some Responses streams emit tool-call argument deltas without stable call id/name.
+	 * Track the last observed tool identity from output_item events so we can still
+	 * emit `tool_call_partial` chunks (tool-call-only streams).
+	 */
+	private pendingToolCallId: string | undefined
+	private pendingToolCallName: string | undefined
 	// Resolved service tier from Responses API (actual tier used by OpenAI)
 	private lastServiceTier: ServiceTier | undefined
 	// Complete response output array (includes reasoning items with encrypted_content)
@@ -51,6 +58,7 @@ export class OpenAiNativeHandler extends BaseProvider implements SingleCompletio
 		"response.reasoning_summary_text.delta",
 		"response.refusal.delta",
 		"response.output_item.added",
+		"response.output_item.done",
 		"response.done",
 		"response.completed",
 		"response.tool_call_arguments.delta",
@@ -155,6 +163,9 @@ export class OpenAiNativeHandler extends BaseProvider implements SingleCompletio
 		this.lastResponseOutput = undefined
 		// Reset last response id for this request
 		this.lastResponseId = undefined
+		// Reset pending tool identity for this request
+		this.pendingToolCallId = undefined
+		this.pendingToolCallName = undefined
 
 		// Use Responses API for ALL models
 		const { verbosity, reasoning } = this.getModel()
@@ -1136,17 +1147,22 @@ export class OpenAiNativeHandler extends BaseProvider implements SingleCompletio
 			event?.type === "response.tool_call_arguments.delta" ||
 			event?.type === "response.function_call_arguments.delta"
 		) {
-			// Emit partial chunks directly - NativeToolCallParser handles state management
-			const callId = event.call_id || event.tool_call_id || event.id
-			const name = event.name || event.function_name
+			// Some streams omit stable identity on delta events; fall back to the
+			// most recently observed tool identity from output_item events.
+			const callId = event.call_id || event.tool_call_id || event.id || this.pendingToolCallId || undefined
+			const name = event.name || event.function_name || this.pendingToolCallName || undefined
 			const args = event.delta || event.arguments
 
-			yield {
-				type: "tool_call_partial",
-				index: event.index ?? 0,
-				id: callId,
-				name,
-				arguments: args,
+			// Avoid emitting incomplete tool_call_partial chunks; the downstream
+			// NativeToolCallParser needs a name to start a call.
+			if (typeof name === "string" && name.length > 0 && typeof callId === "string" && callId.length > 0) {
+				yield {
+					type: "tool_call_partial",
+					index: event.index ?? 0,
+					id: callId,
+					name,
+					arguments: args,
+				}
 			}
 			return
 		}
@@ -1164,6 +1180,16 @@ export class OpenAiNativeHandler extends BaseProvider implements SingleCompletio
 		if (event?.type === "response.output_item.added" || event?.type === "response.output_item.done") {
 			const item = event?.item
 			if (item) {
+				// Capture tool identity so subsequent argument deltas can be attributed.
+				if (item.type === "function_call" || item.type === "tool_call") {
+					const callId = item.call_id || item.tool_call_id || item.id
+					const name = item.name || item.function?.name || item.function_name
+					if (typeof callId === "string" && callId.length > 0) {
+						this.pendingToolCallId = callId
+						this.pendingToolCallName = typeof name === "string" ? name : undefined
+					}
+				}
+
 				if (item.type === "text" && item.text) {
 					yield { type: "text", text: item.text }
 				} else if (item.type === "reasoning" && item.text) {