Skip to content

OpenAI Chat Completions model

OpenAIChatCompletionsModel

Bases: Model

Source code in src/agents/models/openai_chatcompletions.py
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
class OpenAIChatCompletionsModel(Model):
    def __init__(
        self,
        model: str | ChatModel,
        openai_client: AsyncOpenAI,
    ) -> None:
        self.model = model
        self._client = openai_client

    def _non_null_or_not_given(self, value: Any) -> Any:
        return value if value is not None else NOT_GIVEN

    async def get_response(
        self,
        system_instructions: str | None,
        input: str | list[TResponseInputItem],
        model_settings: ModelSettings,
        tools: list[Tool],
        output_schema: AgentOutputSchema | None,
        handoffs: list[Handoff],
        tracing: ModelTracing,
        previous_response_id: str | None,
    ) -> ModelResponse:
        with generation_span(
            model=str(self.model),
            model_config=dataclasses.asdict(model_settings)
            | {"base_url": str(self._client.base_url)},
            disabled=tracing.is_disabled(),
        ) as span_generation:
            response = await self._fetch_response(
                system_instructions,
                input,
                model_settings,
                tools,
                output_schema,
                handoffs,
                span_generation,
                tracing,
                stream=False,
            )

            if _debug.DONT_LOG_MODEL_DATA:
                logger.debug("Received model response")
            else:
                logger.debug(
                    f"LLM resp:\n{json.dumps(response.choices[0].message.model_dump(), indent=2)}\n"
                )

            usage = (
                Usage(
                    requests=1,
                    input_tokens=response.usage.prompt_tokens,
                    output_tokens=response.usage.completion_tokens,
                    total_tokens=response.usage.total_tokens,
                )
                if response.usage
                else Usage()
            )
            if tracing.include_data():
                span_generation.span_data.output = [response.choices[0].message.model_dump()]
            span_generation.span_data.usage = {
                "input_tokens": usage.input_tokens,
                "output_tokens": usage.output_tokens,
            }

            items = _Converter.message_to_output_items(response.choices[0].message)

            return ModelResponse(
                output=items,
                usage=usage,
                response_id=None,
            )

    async def stream_response(
        self,
        system_instructions: str | None,
        input: str | list[TResponseInputItem],
        model_settings: ModelSettings,
        tools: list[Tool],
        output_schema: AgentOutputSchema | None,
        handoffs: list[Handoff],
        tracing: ModelTracing,
        *,
        previous_response_id: str | None,
    ) -> AsyncIterator[TResponseStreamEvent]:
        """
        Yields a partial message as it is generated, as well as the usage information.
        """
        with generation_span(
            model=str(self.model),
            model_config=dataclasses.asdict(model_settings)
            | {"base_url": str(self._client.base_url)},
            disabled=tracing.is_disabled(),
        ) as span_generation:
            response, stream = await self._fetch_response(
                system_instructions,
                input,
                model_settings,
                tools,
                output_schema,
                handoffs,
                span_generation,
                tracing,
                stream=True,
            )

            usage: CompletionUsage | None = None
            state = _StreamingState()

            async for chunk in stream:
                if not state.started:
                    state.started = True
                    yield ResponseCreatedEvent(
                        response=response,
                        type="response.created",
                    )

                # The usage is only available in the last chunk
                usage = chunk.usage

                if not chunk.choices or not chunk.choices[0].delta:
                    continue

                delta = chunk.choices[0].delta

                # Handle text
                if delta.content:
                    if not state.text_content_index_and_output:
                        # Initialize a content tracker for streaming text
                        state.text_content_index_and_output = (
                            0 if not state.refusal_content_index_and_output else 1,
                            ResponseOutputText(
                                text="",
                                type="output_text",
                                annotations=[],
                            ),
                        )
                        # Start a new assistant message stream
                        assistant_item = ResponseOutputMessage(
                            id=FAKE_RESPONSES_ID,
                            content=[],
                            role="assistant",
                            type="message",
                            status="in_progress",
                        )
                        # Notify consumers of the start of a new output message + first content part
                        yield ResponseOutputItemAddedEvent(
                            item=assistant_item,
                            output_index=0,
                            type="response.output_item.added",
                        )
                        yield ResponseContentPartAddedEvent(
                            content_index=state.text_content_index_and_output[0],
                            item_id=FAKE_RESPONSES_ID,
                            output_index=0,
                            part=ResponseOutputText(
                                text="",
                                type="output_text",
                                annotations=[],
                            ),
                            type="response.content_part.added",
                        )
                    # Emit the delta for this segment of content
                    yield ResponseTextDeltaEvent(
                        content_index=state.text_content_index_and_output[0],
                        delta=delta.content,
                        item_id=FAKE_RESPONSES_ID,
                        output_index=0,
                        type="response.output_text.delta",
                    )
                    # Accumulate the text into the response part
                    state.text_content_index_and_output[1].text += delta.content

                # Handle refusals (model declines to answer)
                if delta.refusal:
                    if not state.refusal_content_index_and_output:
                        # Initialize a content tracker for streaming refusal text
                        state.refusal_content_index_and_output = (
                            0 if not state.text_content_index_and_output else 1,
                            ResponseOutputRefusal(refusal="", type="refusal"),
                        )
                        # Start a new assistant message if one doesn't exist yet (in-progress)
                        assistant_item = ResponseOutputMessage(
                            id=FAKE_RESPONSES_ID,
                            content=[],
                            role="assistant",
                            type="message",
                            status="in_progress",
                        )
                        # Notify downstream that assistant message + first content part are starting
                        yield ResponseOutputItemAddedEvent(
                            item=assistant_item,
                            output_index=0,
                            type="response.output_item.added",
                        )
                        yield ResponseContentPartAddedEvent(
                            content_index=state.refusal_content_index_and_output[0],
                            item_id=FAKE_RESPONSES_ID,
                            output_index=0,
                            part=ResponseOutputText(
                                text="",
                                type="output_text",
                                annotations=[],
                            ),
                            type="response.content_part.added",
                        )
                    # Emit the delta for this segment of refusal
                    yield ResponseRefusalDeltaEvent(
                        content_index=state.refusal_content_index_and_output[0],
                        delta=delta.refusal,
                        item_id=FAKE_RESPONSES_ID,
                        output_index=0,
                        type="response.refusal.delta",
                    )
                    # Accumulate the refusal string in the output part
                    state.refusal_content_index_and_output[1].refusal += delta.refusal

                # Handle tool calls
                # Because we don't know the name of the function until the end of the stream, we'll
                # save everything and yield events at the end
                if delta.tool_calls:
                    for tc_delta in delta.tool_calls:
                        if tc_delta.index not in state.function_calls:
                            state.function_calls[tc_delta.index] = ResponseFunctionToolCall(
                                id=FAKE_RESPONSES_ID,
                                arguments="",
                                name="",
                                type="function_call",
                                call_id="",
                            )
                        tc_function = tc_delta.function

                        state.function_calls[tc_delta.index].arguments += (
                            tc_function.arguments if tc_function else ""
                        ) or ""
                        state.function_calls[tc_delta.index].name += (
                            tc_function.name if tc_function else ""
                        ) or ""
                        state.function_calls[tc_delta.index].call_id += tc_delta.id or ""

            function_call_starting_index = 0
            if state.text_content_index_and_output:
                function_call_starting_index += 1
                # Send end event for this content part
                yield ResponseContentPartDoneEvent(
                    content_index=state.text_content_index_and_output[0],
                    item_id=FAKE_RESPONSES_ID,
                    output_index=0,
                    part=state.text_content_index_and_output[1],
                    type="response.content_part.done",
                )

            if state.refusal_content_index_and_output:
                function_call_starting_index += 1
                # Send end event for this content part
                yield ResponseContentPartDoneEvent(
                    content_index=state.refusal_content_index_and_output[0],
                    item_id=FAKE_RESPONSES_ID,
                    output_index=0,
                    part=state.refusal_content_index_and_output[1],
                    type="response.content_part.done",
                )

            # Actually send events for the function calls
            for function_call in state.function_calls.values():
                # First, a ResponseOutputItemAdded for the function call
                yield ResponseOutputItemAddedEvent(
                    item=ResponseFunctionToolCall(
                        id=FAKE_RESPONSES_ID,
                        call_id=function_call.call_id,
                        arguments=function_call.arguments,
                        name=function_call.name,
                        type="function_call",
                    ),
                    output_index=function_call_starting_index,
                    type="response.output_item.added",
                )
                # Then, yield the args
                yield ResponseFunctionCallArgumentsDeltaEvent(
                    delta=function_call.arguments,
                    item_id=FAKE_RESPONSES_ID,
                    output_index=function_call_starting_index,
                    type="response.function_call_arguments.delta",
                )
                # Finally, the ResponseOutputItemDone
                yield ResponseOutputItemDoneEvent(
                    item=ResponseFunctionToolCall(
                        id=FAKE_RESPONSES_ID,
                        call_id=function_call.call_id,
                        arguments=function_call.arguments,
                        name=function_call.name,
                        type="function_call",
                    ),
                    output_index=function_call_starting_index,
                    type="response.output_item.done",
                )

            # Finally, send the Response completed event
            outputs: list[ResponseOutputItem] = []
            if state.text_content_index_and_output or state.refusal_content_index_and_output:
                assistant_msg = ResponseOutputMessage(
                    id=FAKE_RESPONSES_ID,
                    content=[],
                    role="assistant",
                    type="message",
                    status="completed",
                )
                if state.text_content_index_and_output:
                    assistant_msg.content.append(state.text_content_index_and_output[1])
                if state.refusal_content_index_and_output:
                    assistant_msg.content.append(state.refusal_content_index_and_output[1])
                outputs.append(assistant_msg)

                # send a ResponseOutputItemDone for the assistant message
                yield ResponseOutputItemDoneEvent(
                    item=assistant_msg,
                    output_index=0,
                    type="response.output_item.done",
                )

            for function_call in state.function_calls.values():
                outputs.append(function_call)

            final_response = response.model_copy()
            final_response.output = outputs
            final_response.usage = (
                ResponseUsage(
                    input_tokens=usage.prompt_tokens,
                    output_tokens=usage.completion_tokens,
                    total_tokens=usage.total_tokens,
                    output_tokens_details=OutputTokensDetails(
                        reasoning_tokens=usage.completion_tokens_details.reasoning_tokens
                        if usage.completion_tokens_details
                        and usage.completion_tokens_details.reasoning_tokens
                        else 0
                    ),
                    input_tokens_details=InputTokensDetails(
                        cached_tokens=usage.prompt_tokens_details.cached_tokens
                        if usage.prompt_tokens_details and usage.prompt_tokens_details.cached_tokens
                        else 0
                    ),
                )
                if usage
                else None
            )

            yield ResponseCompletedEvent(
                response=final_response,
                type="response.completed",
            )
            if tracing.include_data():
                span_generation.span_data.output = [final_response.model_dump()]

            if usage:
                span_generation.span_data.usage = {
                    "input_tokens": usage.prompt_tokens,
                    "output_tokens": usage.completion_tokens,
                }

    @overload
    async def _fetch_response(
        self,
        system_instructions: str | None,
        input: str | list[TResponseInputItem],
        model_settings: ModelSettings,
        tools: list[Tool],
        output_schema: AgentOutputSchema | None,
        handoffs: list[Handoff],
        span: Span[GenerationSpanData],
        tracing: ModelTracing,
        stream: Literal[True],
    ) -> tuple[Response, AsyncStream[ChatCompletionChunk]]: ...

    @overload
    async def _fetch_response(
        self,
        system_instructions: str | None,
        input: str | list[TResponseInputItem],
        model_settings: ModelSettings,
        tools: list[Tool],
        output_schema: AgentOutputSchema | None,
        handoffs: list[Handoff],
        span: Span[GenerationSpanData],
        tracing: ModelTracing,
        stream: Literal[False],
    ) -> ChatCompletion: ...

    async def _fetch_response(
        self,
        system_instructions: str | None,
        input: str | list[TResponseInputItem],
        model_settings: ModelSettings,
        tools: list[Tool],
        output_schema: AgentOutputSchema | None,
        handoffs: list[Handoff],
        span: Span[GenerationSpanData],
        tracing: ModelTracing,
        stream: bool = False,
    ) -> ChatCompletion | tuple[Response, AsyncStream[ChatCompletionChunk]]:
        converted_messages = _Converter.items_to_messages(input)

        if system_instructions:
            converted_messages.insert(
                0,
                {
                    "content": system_instructions,
                    "role": "system",
                },
            )
        if tracing.include_data():
            span.span_data.input = converted_messages

        parallel_tool_calls = (
            True
            if model_settings.parallel_tool_calls and tools and len(tools) > 0
            else False
            if model_settings.parallel_tool_calls is False
            else NOT_GIVEN
        )
        tool_choice = _Converter.convert_tool_choice(model_settings.tool_choice)
        response_format = _Converter.convert_response_format(output_schema)

        converted_tools = [ToolConverter.to_openai(tool) for tool in tools] if tools else []

        for handoff in handoffs:
            converted_tools.append(ToolConverter.convert_handoff_tool(handoff))

        if _debug.DONT_LOG_MODEL_DATA:
            logger.debug("Calling LLM")
        else:
            logger.debug(
                f"{json.dumps(converted_messages, indent=2)}\n"
                f"Tools:\n{json.dumps(converted_tools, indent=2)}\n"
                f"Stream: {stream}\n"
                f"Tool choice: {tool_choice}\n"
                f"Response format: {response_format}\n"
            )

        reasoning_effort = model_settings.reasoning.effort if model_settings.reasoning else None
        store = _Converter.get_store_param(self._get_client(), model_settings)

        stream_options = _Converter.get_stream_options_param(
            self._get_client(), model_settings, stream=stream
        )

        ret = await self._get_client().chat.completions.create(
            model=self.model,
            messages=converted_messages,
            tools=converted_tools or NOT_GIVEN,
            temperature=self._non_null_or_not_given(model_settings.temperature),
            top_p=self._non_null_or_not_given(model_settings.top_p),
            frequency_penalty=self._non_null_or_not_given(model_settings.frequency_penalty),
            presence_penalty=self._non_null_or_not_given(model_settings.presence_penalty),
            max_tokens=self._non_null_or_not_given(model_settings.max_tokens),
            tool_choice=tool_choice,
            response_format=response_format,
            parallel_tool_calls=parallel_tool_calls,
            stream=stream,
            stream_options=self._non_null_or_not_given(stream_options),
            store=self._non_null_or_not_given(store),
            reasoning_effort=self._non_null_or_not_given(reasoning_effort),
            extra_headers=_HEADERS,
            extra_query=model_settings.extra_query,
            extra_body=model_settings.extra_body,
            metadata=self._non_null_or_not_given(model_settings.metadata),
        )

        if isinstance(ret, ChatCompletion):
            return ret

        response = Response(
            id=FAKE_RESPONSES_ID,
            created_at=time.time(),
            model=self.model,
            object="response",
            output=[],
            tool_choice=cast(Literal["auto", "required", "none"], tool_choice)
            if tool_choice != NOT_GIVEN
            else "auto",
            top_p=model_settings.top_p,
            temperature=model_settings.temperature,
            tools=[],
            parallel_tool_calls=parallel_tool_calls or False,
            reasoning=model_settings.reasoning,
        )
        return response, ret

    def _get_client(self) -> AsyncOpenAI:
        if self._client is None:
            self._client = AsyncOpenAI()
        return self._client

stream_response async

stream_response(
    system_instructions: str | None,
    input: str | list[TResponseInputItem],
    model_settings: ModelSettings,
    tools: list[Tool],
    output_schema: AgentOutputSchema | None,
    handoffs: list[Handoff],
    tracing: ModelTracing,
    *,
    previous_response_id: str | None,
) -> AsyncIterator[TResponseStreamEvent]

Yields a partial message as it is generated, as well as the usage information.

Source code in src/agents/models/openai_chatcompletions.py
async def stream_response(
    self,
    system_instructions: str | None,
    input: str | list[TResponseInputItem],
    model_settings: ModelSettings,
    tools: list[Tool],
    output_schema: AgentOutputSchema | None,
    handoffs: list[Handoff],
    tracing: ModelTracing,
    *,
    previous_response_id: str | None,
) -> AsyncIterator[TResponseStreamEvent]:
    """
    Yields a partial message as it is generated, as well as the usage information.
    """
    with generation_span(
        model=str(self.model),
        model_config=dataclasses.asdict(model_settings)
        | {"base_url": str(self._client.base_url)},
        disabled=tracing.is_disabled(),
    ) as span_generation:
        response, stream = await self._fetch_response(
            system_instructions,
            input,
            model_settings,
            tools,
            output_schema,
            handoffs,
            span_generation,
            tracing,
            stream=True,
        )

        usage: CompletionUsage | None = None
        state = _StreamingState()

        async for chunk in stream:
            if not state.started:
                state.started = True
                yield ResponseCreatedEvent(
                    response=response,
                    type="response.created",
                )

            # The usage is only available in the last chunk
            usage = chunk.usage

            if not chunk.choices or not chunk.choices[0].delta:
                continue

            delta = chunk.choices[0].delta

            # Handle text
            if delta.content:
                if not state.text_content_index_and_output:
                    # Initialize a content tracker for streaming text
                    state.text_content_index_and_output = (
                        0 if not state.refusal_content_index_and_output else 1,
                        ResponseOutputText(
                            text="",
                            type="output_text",
                            annotations=[],
                        ),
                    )
                    # Start a new assistant message stream
                    assistant_item = ResponseOutputMessage(
                        id=FAKE_RESPONSES_ID,
                        content=[],
                        role="assistant",
                        type="message",
                        status="in_progress",
                    )
                    # Notify consumers of the start of a new output message + first content part
                    yield ResponseOutputItemAddedEvent(
                        item=assistant_item,
                        output_index=0,
                        type="response.output_item.added",
                    )
                    yield ResponseContentPartAddedEvent(
                        content_index=state.text_content_index_and_output[0],
                        item_id=FAKE_RESPONSES_ID,
                        output_index=0,
                        part=ResponseOutputText(
                            text="",
                            type="output_text",
                            annotations=[],
                        ),
                        type="response.content_part.added",
                    )
                # Emit the delta for this segment of content
                yield ResponseTextDeltaEvent(
                    content_index=state.text_content_index_and_output[0],
                    delta=delta.content,
                    item_id=FAKE_RESPONSES_ID,
                    output_index=0,
                    type="response.output_text.delta",
                )
                # Accumulate the text into the response part
                state.text_content_index_and_output[1].text += delta.content

            # Handle refusals (model declines to answer)
            if delta.refusal:
                if not state.refusal_content_index_and_output:
                    # Initialize a content tracker for streaming refusal text
                    state.refusal_content_index_and_output = (
                        0 if not state.text_content_index_and_output else 1,
                        ResponseOutputRefusal(refusal="", type="refusal"),
                    )
                    # Start a new assistant message if one doesn't exist yet (in-progress)
                    assistant_item = ResponseOutputMessage(
                        id=FAKE_RESPONSES_ID,
                        content=[],
                        role="assistant",
                        type="message",
                        status="in_progress",
                    )
                    # Notify downstream that assistant message + first content part are starting
                    yield ResponseOutputItemAddedEvent(
                        item=assistant_item,
                        output_index=0,
                        type="response.output_item.added",
                    )
                    yield ResponseContentPartAddedEvent(
                        content_index=state.refusal_content_index_and_output[0],
                        item_id=FAKE_RESPONSES_ID,
                        output_index=0,
                        part=ResponseOutputText(
                            text="",
                            type="output_text",
                            annotations=[],
                        ),
                        type="response.content_part.added",
                    )
                # Emit the delta for this segment of refusal
                yield ResponseRefusalDeltaEvent(
                    content_index=state.refusal_content_index_and_output[0],
                    delta=delta.refusal,
                    item_id=FAKE_RESPONSES_ID,
                    output_index=0,
                    type="response.refusal.delta",
                )
                # Accumulate the refusal string in the output part
                state.refusal_content_index_and_output[1].refusal += delta.refusal

            # Handle tool calls
            # Because we don't know the name of the function until the end of the stream, we'll
            # save everything and yield events at the end
            if delta.tool_calls:
                for tc_delta in delta.tool_calls:
                    if tc_delta.index not in state.function_calls:
                        state.function_calls[tc_delta.index] = ResponseFunctionToolCall(
                            id=FAKE_RESPONSES_ID,
                            arguments="",
                            name="",
                            type="function_call",
                            call_id="",
                        )
                    tc_function = tc_delta.function

                    state.function_calls[tc_delta.index].arguments += (
                        tc_function.arguments if tc_function else ""
                    ) or ""
                    state.function_calls[tc_delta.index].name += (
                        tc_function.name if tc_function else ""
                    ) or ""
                    state.function_calls[tc_delta.index].call_id += tc_delta.id or ""

        function_call_starting_index = 0
        if state.text_content_index_and_output:
            function_call_starting_index += 1
            # Send end event for this content part
            yield ResponseContentPartDoneEvent(
                content_index=state.text_content_index_and_output[0],
                item_id=FAKE_RESPONSES_ID,
                output_index=0,
                part=state.text_content_index_and_output[1],
                type="response.content_part.done",
            )

        if state.refusal_content_index_and_output:
            function_call_starting_index += 1
            # Send end event for this content part
            yield ResponseContentPartDoneEvent(
                content_index=state.refusal_content_index_and_output[0],
                item_id=FAKE_RESPONSES_ID,
                output_index=0,
                part=state.refusal_content_index_and_output[1],
                type="response.content_part.done",
            )

        # Actually send events for the function calls
        for function_call in state.function_calls.values():
            # First, a ResponseOutputItemAdded for the function call
            yield ResponseOutputItemAddedEvent(
                item=ResponseFunctionToolCall(
                    id=FAKE_RESPONSES_ID,
                    call_id=function_call.call_id,
                    arguments=function_call.arguments,
                    name=function_call.name,
                    type="function_call",
                ),
                output_index=function_call_starting_index,
                type="response.output_item.added",
            )
            # Then, yield the args
            yield ResponseFunctionCallArgumentsDeltaEvent(
                delta=function_call.arguments,
                item_id=FAKE_RESPONSES_ID,
                output_index=function_call_starting_index,
                type="response.function_call_arguments.delta",
            )
            # Finally, the ResponseOutputItemDone
            yield ResponseOutputItemDoneEvent(
                item=ResponseFunctionToolCall(
                    id=FAKE_RESPONSES_ID,
                    call_id=function_call.call_id,
                    arguments=function_call.arguments,
                    name=function_call.name,
                    type="function_call",
                ),
                output_index=function_call_starting_index,
                type="response.output_item.done",
            )

        # Finally, send the Response completed event
        outputs: list[ResponseOutputItem] = []
        if state.text_content_index_and_output or state.refusal_content_index_and_output:
            assistant_msg = ResponseOutputMessage(
                id=FAKE_RESPONSES_ID,
                content=[],
                role="assistant",
                type="message",
                status="completed",
            )
            if state.text_content_index_and_output:
                assistant_msg.content.append(state.text_content_index_and_output[1])
            if state.refusal_content_index_and_output:
                assistant_msg.content.append(state.refusal_content_index_and_output[1])
            outputs.append(assistant_msg)

            # send a ResponseOutputItemDone for the assistant message
            yield ResponseOutputItemDoneEvent(
                item=assistant_msg,
                output_index=0,
                type="response.output_item.done",
            )

        for function_call in state.function_calls.values():
            outputs.append(function_call)

        final_response = response.model_copy()
        final_response.output = outputs
        final_response.usage = (
            ResponseUsage(
                input_tokens=usage.prompt_tokens,
                output_tokens=usage.completion_tokens,
                total_tokens=usage.total_tokens,
                output_tokens_details=OutputTokensDetails(
                    reasoning_tokens=usage.completion_tokens_details.reasoning_tokens
                    if usage.completion_tokens_details
                    and usage.completion_tokens_details.reasoning_tokens
                    else 0
                ),
                input_tokens_details=InputTokensDetails(
                    cached_tokens=usage.prompt_tokens_details.cached_tokens
                    if usage.prompt_tokens_details and usage.prompt_tokens_details.cached_tokens
                    else 0
                ),
            )
            if usage
            else None
        )

        yield ResponseCompletedEvent(
            response=final_response,
            type="response.completed",
        )
        if tracing.include_data():
            span_generation.span_data.output = [final_response.model_dump()]

        if usage:
            span_generation.span_data.usage = {
                "input_tokens": usage.prompt_tokens,
                "output_tokens": usage.completion_tokens,
            }