sentry-sdk 3.0.0a2__py2.py3-none-any.whl → 3.0.0a4__py2.py3-none-any.whl

sentry_sdk/integrations/openai.py

@@ -1,3 +1,4 @@
+from __future__ import annotations
 from functools import wraps
 
 import sentry_sdk
@@ -10,6 +11,7 @@ from sentry_sdk.scope import should_send_default_pii
 from sentry_sdk.utils import (
     capture_internal_exceptions,
     event_from_exception,
+    safe_serialize,
 )
 
 from typing import TYPE_CHECKING
@@ -19,6 +21,11 @@ if TYPE_CHECKING:
     from sentry_sdk.tracing import Span
 
 try:
+    try:
+        from openai import NOT_GIVEN
+    except ImportError:
+        NOT_GIVEN = None
+
     from openai.resources.chat.completions import Completions, AsyncCompletions
     from openai.resources import Embeddings, AsyncEmbeddings
 
@@ -27,13 +34,24 @@ try:
 except ImportError:
     raise DidNotEnable("OpenAI not installed")
 
+RESPONSES_API_ENABLED = True
+try:
+    # responses API support was introduced in v1.66.0
+    from openai.resources.responses import Responses, AsyncResponses
+    from openai.types.responses.response_completed_event import ResponseCompletedEvent
+except ImportError:
+    RESPONSES_API_ENABLED = False
+
 
 class OpenAIIntegration(Integration):
     identifier = "openai"
     origin = f"auto.ai.{identifier}"
 
-    def __init__(self, include_prompts=True, tiktoken_encoding_name=None):
-        # type: (OpenAIIntegration, bool, Optional[str]) -> None
+    def __init__(
+        self: OpenAIIntegration,
+        include_prompts: bool = True,
+        tiktoken_encoding_name: Optional[str] = None,
+    ) -> None:
         self.include_prompts = include_prompts
 
         self.tiktoken_encoding = None
@@ -43,25 +61,32 @@ class OpenAIIntegration(Integration):
             self.tiktoken_encoding = tiktoken.get_encoding(tiktoken_encoding_name)
 
     @staticmethod
-    def setup_once():
-        # type: () -> None
+    def setup_once() -> None:
         Completions.create = _wrap_chat_completion_create(Completions.create)
-        Embeddings.create = _wrap_embeddings_create(Embeddings.create)
-
         AsyncCompletions.create = _wrap_async_chat_completion_create(
             AsyncCompletions.create
         )
+
+        Embeddings.create = _wrap_embeddings_create(Embeddings.create)
         AsyncEmbeddings.create = _wrap_async_embeddings_create(AsyncEmbeddings.create)
 
-    def count_tokens(self, s):
-        # type: (OpenAIIntegration, str) -> int
+        if RESPONSES_API_ENABLED:
+            Responses.create = _wrap_responses_create(Responses.create)
+            AsyncResponses.create = _wrap_async_responses_create(AsyncResponses.create)
+
+    def count_tokens(self: OpenAIIntegration, s: str) -> int:
         if self.tiktoken_encoding is not None:
             return len(self.tiktoken_encoding.encode_ordinary(s))
         return 0
 
 
-def _capture_exception(exc):
-    # type: (Any) -> None
+def _capture_exception(exc: Any) -> None:
+    # Close an eventually open span
+    # We need to do this by hand because we are not using the start_span context manager
+    current_span = sentry_sdk.get_current_span()
+    if current_span is not None:
+        current_span.__exit__(None, None, None)
+
     event, hint = event_from_exception(
         exc,
         client_options=sentry_sdk.get_client().options,
@@ -70,185 +95,327 @@ def _capture_exception(exc):
     sentry_sdk.capture_event(event, hint=hint)
 
 
-def _calculate_chat_completion_usage(
-    messages, response, span, streaming_message_responses, count_tokens
-):
-    # type: (Iterable[ChatCompletionMessageParam], Any, Span, Optional[List[str]], Callable[..., Any]) -> None
-    completion_tokens = 0  # type: Optional[int]
-    prompt_tokens = 0  # type: Optional[int]
-    total_tokens = 0  # type: Optional[int]
+def _get_usage(usage: Any, names: List[str]) -> int:
+    for name in names:
+        if hasattr(usage, name) and isinstance(getattr(usage, name), int):
+            return getattr(usage, name)
+    return 0
+
+
+def _calculate_token_usage(
+    messages: Optional[Iterable[ChatCompletionMessageParam]],
+    response: Any,
+    span: Span,
+    streaming_message_responses: Optional[List[str]],
+    count_tokens: Callable[..., Any],
+) -> None:
+    input_tokens: Optional[int] = 0
+    input_tokens_cached: Optional[int] = 0
+    output_tokens: Optional[int] = 0
+    output_tokens_reasoning: Optional[int] = 0
+    total_tokens: Optional[int] = 0
+
     if hasattr(response, "usage"):
-        if hasattr(response.usage, "completion_tokens") and isinstance(
-            response.usage.completion_tokens, int
-        ):
-            completion_tokens = response.usage.completion_tokens
-        if hasattr(response.usage, "prompt_tokens") and isinstance(
-            response.usage.prompt_tokens, int
-        ):
-            prompt_tokens = response.usage.prompt_tokens
-        if hasattr(response.usage, "total_tokens") and isinstance(
-            response.usage.total_tokens, int
-        ):
-            total_tokens = response.usage.total_tokens
-
-    if prompt_tokens == 0:
-        for message in messages:
-            if "content" in message:
-                prompt_tokens += count_tokens(message["content"])
-
-    if completion_tokens == 0:
+        input_tokens = _get_usage(response.usage, ["input_tokens", "prompt_tokens"])
+        if hasattr(response.usage, "input_tokens_details"):
+            input_tokens_cached = _get_usage(
+                response.usage.input_tokens_details, ["cached_tokens"]
+            )
+
+        output_tokens = _get_usage(
+            response.usage, ["output_tokens", "completion_tokens"]
+        )
+        if hasattr(response.usage, "output_tokens_details"):
+            output_tokens_reasoning = _get_usage(
+                response.usage.output_tokens_details, ["reasoning_tokens"]
+            )
+
+        total_tokens = _get_usage(response.usage, ["total_tokens"])
+
+    # Manually count tokens
+    if input_tokens == 0:
+        for message in messages or []:
+            if isinstance(message, dict) and "content" in message:
+                input_tokens += count_tokens(message["content"])
+            elif isinstance(message, str):
+                input_tokens += count_tokens(message)
+
+    if output_tokens == 0:
         if streaming_message_responses is not None:
             for message in streaming_message_responses:
-                completion_tokens += count_tokens(message)
+                output_tokens += count_tokens(message)
         elif hasattr(response, "choices"):
             for choice in response.choices:
                 if hasattr(choice, "message"):
-                    completion_tokens += count_tokens(choice.message)
-
-    if prompt_tokens == 0:
-        prompt_tokens = None
-    if completion_tokens == 0:
-        completion_tokens = None
-    if total_tokens == 0:
-        total_tokens = None
-    record_token_usage(span, prompt_tokens, completion_tokens, total_tokens)
-
+                    output_tokens += count_tokens(choice.message)
+
+    # Do not set token data if it is 0
+    input_tokens = input_tokens or None
+    input_tokens_cached = input_tokens_cached or None
+    output_tokens = output_tokens or None
+    output_tokens_reasoning = output_tokens_reasoning or None
+    total_tokens = total_tokens or None
+
+    record_token_usage(
+        span,
+        input_tokens=input_tokens,
+        input_tokens_cached=input_tokens_cached,
+        output_tokens=output_tokens,
+        output_tokens_reasoning=output_tokens_reasoning,
+        total_tokens=total_tokens,
+    )
 
-def _new_chat_completion_common(f, *args, **kwargs):
-    # type: (Any, *Any, **Any) -> Any
-    integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
-    if integration is None:
-        return f(*args, **kwargs)
 
-    if "messages" not in kwargs:
-        # invalid call (in all versions of openai), let it return error
-        return f(*args, **kwargs)
+def _set_input_data(
+    span: Span, kwargs: Any, operation: str, integration: OpenAIIntegration
+) -> None:
+    # Input messages (the prompt or data sent to the model)
+    messages = kwargs.get("messages")
+    if messages is None:
+        messages = kwargs.get("input")
+
+    if isinstance(messages, str):
+        messages = [messages]
+
+    if (
+        messages is not None
+        and len(messages) > 0
+        and should_send_default_pii()
+        and integration.include_prompts
+    ):
+        set_data_normalized(span, SPANDATA.GEN_AI_REQUEST_MESSAGES, messages)
+
+    # Input attributes: Common
+    set_data_normalized(span, SPANDATA.GEN_AI_SYSTEM, "openai")
+    set_data_normalized(span, SPANDATA.GEN_AI_OPERATION_NAME, operation)
+
+    # Input attributes: Optional
+    kwargs_keys_to_attributes = {
+        "model": SPANDATA.GEN_AI_REQUEST_MODEL,
+        "stream": SPANDATA.GEN_AI_RESPONSE_STREAMING,
+        "max_tokens": SPANDATA.GEN_AI_REQUEST_MAX_TOKENS,
+        "presence_penalty": SPANDATA.GEN_AI_REQUEST_PRESENCE_PENALTY,
+        "frequency_penalty": SPANDATA.GEN_AI_REQUEST_FREQUENCY_PENALTY,
+        "temperature": SPANDATA.GEN_AI_REQUEST_TEMPERATURE,
+        "top_p": SPANDATA.GEN_AI_REQUEST_TOP_P,
+    }
+    for key, attribute in kwargs_keys_to_attributes.items():
+        value = kwargs.get(key)
+
+        if value is not NOT_GIVEN and value is not None:
+            set_data_normalized(span, attribute, value)
+
+    # Input attributes: Tools
+    tools = kwargs.get("tools")
+    if tools is not NOT_GIVEN and tools is not None and len(tools) > 0:
+        set_data_normalized(
+            span, SPANDATA.GEN_AI_REQUEST_AVAILABLE_TOOLS, safe_serialize(tools)
+        )
 
-    try:
-        iter(kwargs["messages"])
-    except TypeError:
-        # invalid call (in all versions), messages must be iterable
-        return f(*args, **kwargs)
 
-    kwargs["messages"] = list(kwargs["messages"])
-    messages = kwargs["messages"]
-    model = kwargs.get("model")
-    streaming = kwargs.get("stream")
+def _set_output_data(
+    span: Span,
+    response: Any,
+    kwargs: Any,
+    integration: OpenAIIntegration,
+    finish_span: bool = True,
+) -> None:
+    if hasattr(response, "model"):
+        set_data_normalized(span, SPANDATA.GEN_AI_RESPONSE_MODEL, response.model)
 
-    span = sentry_sdk.start_span(
-        op=consts.OP.OPENAI_CHAT_COMPLETIONS_CREATE,
-        name="Chat Completion",
-        origin=OpenAIIntegration.origin,
-        only_if_parent=True,
-    )
-    span.__enter__()
+    # Input messages (the prompt or data sent to the model)
+    # used for the token usage calculation
+    messages = kwargs.get("messages")
+    if messages is None:
+        messages = kwargs.get("input")
 
-    res = yield f, args, kwargs
+    if messages is not None and isinstance(messages, str):
+        messages = [messages]
 
-    with capture_internal_exceptions():
+    if hasattr(response, "choices"):
         if should_send_default_pii() and integration.include_prompts:
-            set_data_normalized(span, SPANDATA.AI_INPUT_MESSAGES, messages)
-
-        set_data_normalized(span, SPANDATA.AI_MODEL_ID, model)
-        set_data_normalized(span, SPANDATA.AI_STREAMING, streaming)
+            response_text = [choice.message.dict() for choice in response.choices]
+            if len(response_text) > 0:
+                set_data_normalized(
+                    span,
+                    SPANDATA.GEN_AI_RESPONSE_TEXT,
+                    safe_serialize(response_text),
+                )
+        _calculate_token_usage(messages, response, span, None, integration.count_tokens)
+        if finish_span:
+            span.__exit__(None, None, None)
 
-        if hasattr(res, "choices"):
-            if should_send_default_pii() and integration.include_prompts:
+    elif hasattr(response, "output"):
+        if should_send_default_pii() and integration.include_prompts:
+            response_text = [item.to_dict() for item in response.output]
+            if len(response_text) > 0:
                 set_data_normalized(
                     span,
-                    SPANDATA.AI_RESPONSES,
-                    list(map(lambda x: x.message, res.choices)),
+                    SPANDATA.GEN_AI_RESPONSE_TEXT,
+                    safe_serialize(response_text),
                 )
-            _calculate_chat_completion_usage(
-                messages, res, span, None, integration.count_tokens
-            )
+        _calculate_token_usage(messages, response, span, None, integration.count_tokens)
+        if finish_span:
             span.__exit__(None, None, None)
-        elif hasattr(res, "_iterator"):
-            data_buf: list[list[str]] = []  # one for each choice
-
-            old_iterator = res._iterator
-
-            def new_iterator():
-                # type: () -> Iterator[ChatCompletionChunk]
-                with capture_internal_exceptions():
-                    for x in old_iterator:
-                        if hasattr(x, "choices"):
-                            choice_index = 0
-                            for choice in x.choices:
-                                if hasattr(choice, "delta") and hasattr(
-                                    choice.delta, "content"
-                                ):
-                                    content = choice.delta.content
-                                    if len(data_buf) <= choice_index:
-                                        data_buf.append([])
-                                    data_buf[choice_index].append(content or "")
-                                choice_index += 1
-                        yield x
-                    if len(data_buf) > 0:
-                        all_responses = list(
-                            map(lambda chunk: "".join(chunk), data_buf)
+
+    elif hasattr(response, "_iterator"):
+        data_buf: list[list[str]] = []  # one for each choice
+
+        old_iterator = response._iterator
+
+        def new_iterator() -> Iterator[ChatCompletionChunk]:
+            with capture_internal_exceptions():
+                count_tokens_manually = True
+                for x in old_iterator:
+                    # OpenAI chat completion API
+                    if hasattr(x, "choices"):
+                        choice_index = 0
+                        for choice in x.choices:
+                            if hasattr(choice, "delta") and hasattr(
+                                choice.delta, "content"
+                            ):
+                                content = choice.delta.content
+                                if len(data_buf) <= choice_index:
+                                    data_buf.append([])
+                                data_buf[choice_index].append(content or "")
+                            choice_index += 1
+
+                    # OpenAI responses API
+                    elif hasattr(x, "delta"):
+                        if len(data_buf) == 0:
+                            data_buf.append([])
+                        data_buf[0].append(x.delta or "")
+
+                    # OpenAI responses API end of streaming response
+                    if RESPONSES_API_ENABLED and isinstance(x, ResponseCompletedEvent):
+                        _calculate_token_usage(
+                            messages,
+                            x.response,
+                            span,
+                            None,
+                            integration.count_tokens,
                         )
-                        if should_send_default_pii() and integration.include_prompts:
-                            set_data_normalized(
-                                span, SPANDATA.AI_RESPONSES, all_responses
-                            )
-                        _calculate_chat_completion_usage(
+                        count_tokens_manually = False
+
+                    yield x
+
+                if len(data_buf) > 0:
+                    all_responses = ["".join(chunk) for chunk in data_buf]
+                    if should_send_default_pii() and integration.include_prompts:
+                        set_data_normalized(
+                            span, SPANDATA.GEN_AI_RESPONSE_TEXT, all_responses
+                        )
+                    if count_tokens_manually:
+                        _calculate_token_usage(
                             messages,
-                            res,
+                            response,
                             span,
                             all_responses,
                             integration.count_tokens,
                         )
+
+            if finish_span:
                 span.__exit__(None, None, None)
 
-            async def new_iterator_async():
-                # type: () -> AsyncIterator[ChatCompletionChunk]
-                with capture_internal_exceptions():
-                    async for x in old_iterator:
-                        if hasattr(x, "choices"):
-                            choice_index = 0
-                            for choice in x.choices:
-                                if hasattr(choice, "delta") and hasattr(
-                                    choice.delta, "content"
-                                ):
-                                    content = choice.delta.content
-                                    if len(data_buf) <= choice_index:
-                                        data_buf.append([])
-                                    data_buf[choice_index].append(content or "")
-                                choice_index += 1
-                        yield x
-                    if len(data_buf) > 0:
-                        all_responses = list(
-                            map(lambda chunk: "".join(chunk), data_buf)
+        async def new_iterator_async() -> AsyncIterator[ChatCompletionChunk]:
+            with capture_internal_exceptions():
+                count_tokens_manually = True
+                async for x in old_iterator:
+                    # OpenAI chat completion API
+                    if hasattr(x, "choices"):
+                        choice_index = 0
+                        for choice in x.choices:
+                            if hasattr(choice, "delta") and hasattr(
+                                choice.delta, "content"
+                            ):
+                                content = choice.delta.content
+                                if len(data_buf) <= choice_index:
+                                    data_buf.append([])
+                                data_buf[choice_index].append(content or "")
+                            choice_index += 1
+
+                    # OpenAI responses API
+                    elif hasattr(x, "delta"):
+                        if len(data_buf) == 0:
+                            data_buf.append([])
+                        data_buf[0].append(x.delta or "")
+
+                    # OpenAI responses API end of streaming response
+                    if RESPONSES_API_ENABLED and isinstance(x, ResponseCompletedEvent):
+                        _calculate_token_usage(
+                            messages,
+                            x.response,
+                            span,
+                            None,
+                            integration.count_tokens,
+                        )
+                        count_tokens_manually = False
+
+                    yield x
+
+                if len(data_buf) > 0:
+                    all_responses = ["".join(chunk) for chunk in data_buf]
+                    if should_send_default_pii() and integration.include_prompts:
+                        set_data_normalized(
+                            span, SPANDATA.GEN_AI_RESPONSE_TEXT, all_responses
                         )
-                        if should_send_default_pii() and integration.include_prompts:
-                            set_data_normalized(
-                                span, SPANDATA.AI_RESPONSES, all_responses
-                            )
-                        _calculate_chat_completion_usage(
+                    if count_tokens_manually:
+                        _calculate_token_usage(
                             messages,
-                            res,
+                            response,
                             span,
                             all_responses,
                             integration.count_tokens,
                         )
+            if finish_span:
                 span.__exit__(None, None, None)
 
-            if str(type(res._iterator)) == "<class 'async_generator'>":
-                res._iterator = new_iterator_async()
-            else:
-                res._iterator = new_iterator()
-
+        if str(type(response._iterator)) == "<class 'async_generator'>":
+            response._iterator = new_iterator_async()
         else:
-            set_data_normalized(span, "unknown_response", True)
+            response._iterator = new_iterator()
+    else:
+        _calculate_token_usage(messages, response, span, None, integration.count_tokens)
+        if finish_span:
             span.__exit__(None, None, None)
-    return res
 
 
-def _wrap_chat_completion_create(f):
-    # type: (Callable[..., Any]) -> Callable[..., Any]
-    def _execute_sync(f, *args, **kwargs):
-        # type: (Any, *Any, **Any) -> Any
+def _new_chat_completion_common(f: Any, *args: Any, **kwargs: Any) -> Any:
+    integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
+    if integration is None:
+        return f(*args, **kwargs)
+
+    if "messages" not in kwargs:
+        # invalid call (in all versions of openai), let it return error
+        return f(*args, **kwargs)
+
+    try:
+        iter(kwargs["messages"])
+    except TypeError:
+        # invalid call (in all versions), messages must be iterable
+        return f(*args, **kwargs)
+
+    model = kwargs.get("model")
+    operation = "chat"
+
+    span = sentry_sdk.start_span(
+        op=consts.OP.GEN_AI_CHAT,
+        name=f"{operation} {model}",
+        origin=OpenAIIntegration.origin,
+    )
+    span.__enter__()
+
+    _set_input_data(span, kwargs, operation, integration)
+
+    response = yield f, args, kwargs
+
+    _set_output_data(span, response, kwargs, integration, finish_span=True)
+
+    return response
+
+
+def _wrap_chat_completion_create(f: Callable[..., Any]) -> Callable[..., Any]:
+    def _execute_sync(f: Any, *args: Any, **kwargs: Any) -> Any:
         gen = _new_chat_completion_common(f, *args, **kwargs)
 
         try:
@@ -268,8 +435,7 @@ def _wrap_chat_completion_create(f):
             return e.value
 
     @wraps(f)
-    def _sentry_patched_create_sync(*args, **kwargs):
-        # type: (*Any, **Any) -> Any
+    def _sentry_patched_create_sync(*args: Any, **kwargs: Any) -> Any:
         integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
         if integration is None or "messages" not in kwargs:
             # no "messages" means invalid call (in all versions of openai), let it return error
@@ -280,10 +446,8 @@ def _wrap_chat_completion_create(f):
     return _sentry_patched_create_sync
 
 
-def _wrap_async_chat_completion_create(f):
-    # type: (Callable[..., Any]) -> Callable[..., Any]
-    async def _execute_async(f, *args, **kwargs):
-        # type: (Any, *Any, **Any) -> Any
+def _wrap_async_chat_completion_create(f: Callable[..., Any]) -> Callable[..., Any]:
+    async def _execute_async(f: Any, *args: Any, **kwargs: Any) -> Any:
         gen = _new_chat_completion_common(f, *args, **kwargs)
 
         try:
@@ -303,8 +467,7 @@ def _wrap_async_chat_completion_create(f):
             return e.value
 
     @wraps(f)
-    async def _sentry_patched_create_async(*args, **kwargs):
-        # type: (*Any, **Any) -> Any
+    async def _sentry_patched_create_async(*args: Any, **kwargs: Any) -> Any:
         integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
         if integration is None or "messages" not in kwargs:
             # no "messages" means invalid call (in all versions of openai), let it return error
@@ -315,58 +478,30 @@ def _wrap_async_chat_completion_create(f):
     return _sentry_patched_create_async
 
 
-def _new_embeddings_create_common(f, *args, **kwargs):
-    # type: (Any, *Any, **Any) -> Any
+def _new_embeddings_create_common(f: Any, *args: Any, **kwargs: Any) -> Any:
     integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
     if integration is None:
         return f(*args, **kwargs)
 
+    model = kwargs.get("model")
+    operation = "embeddings"
+
     with sentry_sdk.start_span(
-        op=consts.OP.OPENAI_EMBEDDINGS_CREATE,
-        description="OpenAI Embedding Creation",
+        op=consts.OP.GEN_AI_EMBEDDINGS,
+        name=f"{operation} {model}",
         origin=OpenAIIntegration.origin,
-        only_if_parent=True,
     ) as span:
-        if "input" in kwargs and (
-            should_send_default_pii() and integration.include_prompts
-        ):
-            if isinstance(kwargs["input"], str):
-                set_data_normalized(span, SPANDATA.AI_INPUT_MESSAGES, [kwargs["input"]])
-            elif (
-                isinstance(kwargs["input"], list)
-                and len(kwargs["input"]) > 0
-                and isinstance(kwargs["input"][0], str)
-            ):
-                set_data_normalized(span, SPANDATA.AI_INPUT_MESSAGES, kwargs["input"])
-        if "model" in kwargs:
-            set_data_normalized(span, SPANDATA.AI_MODEL_ID, kwargs["model"])
+        _set_input_data(span, kwargs, operation, integration)
 
         response = yield f, args, kwargs
 
-        prompt_tokens = 0
-        total_tokens = 0
-        if hasattr(response, "usage"):
-            if hasattr(response.usage, "prompt_tokens") and isinstance(
-                response.usage.prompt_tokens, int
-            ):
-                prompt_tokens = response.usage.prompt_tokens
-            if hasattr(response.usage, "total_tokens") and isinstance(
-                response.usage.total_tokens, int
-            ):
-                total_tokens = response.usage.total_tokens
-
-        if prompt_tokens == 0:
-            prompt_tokens = integration.count_tokens(kwargs["input"] or "")
-
-        record_token_usage(span, prompt_tokens, None, total_tokens or prompt_tokens)
+        _set_output_data(span, response, kwargs, integration, finish_span=False)
 
         return response
 
 
-def _wrap_embeddings_create(f):
-    # type: (Any) -> Any
-    def _execute_sync(f, *args, **kwargs):
-        # type: (Any, *Any, **Any) -> Any
+def _wrap_embeddings_create(f: Any) -> Any:
+    def _execute_sync(f: Any, *args: Any, **kwargs: Any) -> Any:
         gen = _new_embeddings_create_common(f, *args, **kwargs)
 
         try:
@@ -386,8 +521,7 @@ def _wrap_embeddings_create(f):
             return e.value
 
     @wraps(f)
-    def _sentry_patched_create_sync(*args, **kwargs):
-        # type: (*Any, **Any) -> Any
+    def _sentry_patched_create_sync(*args: Any, **kwargs: Any) -> Any:
         integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
         if integration is None:
             return f(*args, **kwargs)
@@ -397,10 +531,8 @@ def _wrap_embeddings_create(f):
     return _sentry_patched_create_sync
 
 
-def _wrap_async_embeddings_create(f):
-    # type: (Any) -> Any
-    async def _execute_async(f, *args, **kwargs):
-        # type: (Any, *Any, **Any) -> Any
+def _wrap_async_embeddings_create(f: Any) -> Any:
+    async def _execute_async(f: Any, *args: Any, **kwargs: Any) -> Any:
         gen = _new_embeddings_create_common(f, *args, **kwargs)
 
         try:
@@ -420,8 +552,7 @@ def _wrap_async_embeddings_create(f):
             return e.value
 
     @wraps(f)
-    async def _sentry_patched_create_async(*args, **kwargs):
-        # type: (*Any, **Any) -> Any
+    async def _sentry_patched_create_async(*args: Any, **kwargs: Any) -> Any:
         integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
         if integration is None:
             return await f(*args, **kwargs)
@@ -429,3 +560,89 @@ def _wrap_async_embeddings_create(f):
         return await _execute_async(f, *args, **kwargs)
 
     return _sentry_patched_create_async
+
+
+def _new_responses_create_common(f: Any, *args: Any, **kwargs: Any) -> Any:
+    integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
+    if integration is None:
+        return f(*args, **kwargs)
+
+    model = kwargs.get("model")
+    operation = "responses"
+
+    span = sentry_sdk.start_span(
+        op=consts.OP.GEN_AI_RESPONSES,
+        name=f"{operation} {model}",
+        origin=OpenAIIntegration.origin,
+    )
+    span.__enter__()
+
+    _set_input_data(span, kwargs, operation, integration)
+
+    response = yield f, args, kwargs
+
+    _set_output_data(span, response, kwargs, integration, finish_span=True)
+
+    return response
+
+
+def _wrap_responses_create(f: Any) -> Any:
+    def _execute_sync(f: Any, *args: Any, **kwargs: Any) -> Any:
+        gen = _new_responses_create_common(f, *args, **kwargs)
+
+        try:
+            f, args, kwargs = next(gen)
+        except StopIteration as e:
+            return e.value
+
+        try:
+            try:
+                result = f(*args, **kwargs)
+            except Exception as e:
+                _capture_exception(e)
+                raise e from None
+
+            return gen.send(result)
+        except StopIteration as e:
+            return e.value
+
+    @wraps(f)
+    def _sentry_patched_create_sync(*args: Any, **kwargs: Any) -> Any:
+        integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
+        if integration is None:
+            return f(*args, **kwargs)
+
+        return _execute_sync(f, *args, **kwargs)
+
+    return _sentry_patched_create_sync
+
+
+def _wrap_async_responses_create(f: Any) -> Any:
+    async def _execute_async(f: Any, *args: Any, **kwargs: Any) -> Any:
+        gen = _new_responses_create_common(f, *args, **kwargs)
+
+        try:
+            f, args, kwargs = next(gen)
+        except StopIteration as e:
+            return await e.value
+
+        try:
+            try:
+                result = await f(*args, **kwargs)
+            except Exception as e:
+                _capture_exception(e)
+                raise e from None
+
+            return gen.send(result)
+        except StopIteration as e:
+            return e.value
+
+    @wraps(f)
+    async def _sentry_patched_responses_async(*args: Any, **kwargs: Any) -> Any:
+        integration = sentry_sdk.get_client().get_integration(OpenAIIntegration)
+        if integration is None:
+            return await f(*args, **kwargs)
+
+        return await _execute_async(f, *args, **kwargs)
+
+    return _sentry_patched_responses_async