Fixing the issue? Andrew's suggestion

Author: alice-cohere

src/cohere/base_client.py

@@ -1593,7 +1593,8 @@ class AsyncBaseCohere:
         The timeout to be used, in seconds, for requests. By default the timeout is 300 seconds, unless a custom httpx client is used, in which case this default is not enforced.
 
     follow_redirects : typing.Optional[bool]
-        Whether the default httpx client follows redirects or not, this is irrelevant if a custom httpx client is passed in.
+        Whether the async client follows HTTP redirects. Defaults to True. Passed as allow_redirects
+        on each request; does NOT affect TCP connection reuse (connections are always pooled).
 
     httpx_client : typing.Optional[httpx.AsyncClient]
         The httpx client to use for making requests, a preconfigured client is used by default, however this is useful should you want to pass in any custom httpx configuration.
@@ -1625,19 +1626,25 @@ def __init__(
         if token is None:
             raise ApiError(body="The client must be instantiated be either passing in token or setting CO_API_KEY")
 
-        # Create aiohttp session if not provided
+        # Create aiohttp session if not provided.
+        # NOTE: force_close is intentionally NOT derived from follow_redirects.
+        # force_close controls TCP connection reuse (keep-alive pooling); setting it True
+        # causes every request to open and close a fresh TCP socket, exhausting the ephemeral
+        # port range (TIME_WAIT) when making thousands of concurrent calls.
+        # Redirect behaviour is handled per-request via allow_redirects instead.
         if aiohttp_session is None:
             timeout_config = aiohttp.ClientTimeout(total=_defaulted_timeout)
-            connector = aiohttp.TCPConnector(force_close=not follow_redirects) if follow_redirects is not None else aiohttp.TCPConnector()
+            connector = aiohttp.TCPConnector()
             aiohttp_session = aiohttp.ClientSession(timeout=timeout_config, connector=connector)
-        
+
         self._client_wrapper = AsyncClientWrapper(
             base_url=_get_base_url(base_url=base_url, environment=environment),
             client_name=client_name,
             token=token,
             headers=headers,
             aiohttp_session=aiohttp_session,
             timeout=_defaulted_timeout,
+            follow_redirects=follow_redirects if follow_redirects is not None else True,
         )
         self._raw_client = AsyncRawBaseCohere(client_wrapper=self._client_wrapper)
         self._v2: typing.Optional[AsyncV2Client] = None

src/cohere/core/client_wrapper.py

@@ -87,6 +87,7 @@ def __init__(
         timeout: typing.Optional[float] = None,
         async_token: typing.Optional[typing.Callable[[], typing.Awaitable[str]]] = None,
         aiohttp_session: aiohttp.ClientSession,
+        follow_redirects: bool = True,
     ):
         super().__init__(client_name=client_name, token=token, headers=headers, base_url=base_url, timeout=timeout)
         self._async_token = async_token
@@ -96,6 +97,7 @@ def __init__(
             base_timeout=self.get_timeout,
             base_url=self.get_base_url,
             async_base_headers=self.async_get_headers,
+            follow_redirects=follow_redirects,
         )
 
     async def async_get_headers(self) -> typing.Dict[str, str]:

src/cohere/core/http_client.py

@@ -460,12 +460,14 @@ def __init__(
         base_headers: typing.Callable[[], typing.Dict[str, str]],
         base_url: typing.Optional[typing.Callable[[], str]] = None,
         async_base_headers: typing.Optional[typing.Callable[[], typing.Awaitable[typing.Dict[str, str]]]] = None,
+        follow_redirects: bool = True,
     ):
         self.base_url = base_url
         self.base_timeout = base_timeout
         self.base_headers = base_headers
         self.async_base_headers = async_base_headers
         self.aiohttp_session = aiohttp_session
+        self.follow_redirects = follow_redirects
 
     async def _get_headers(self) -> typing.Dict[str, str]:
         if self.async_base_headers is not None:
@@ -570,6 +572,7 @@ async def request(
             json=json_body,
             data=request_data,
             timeout=timeout,
+            allow_redirects=self.follow_redirects,
         )
 
         # Read response body and parse JSON for aiohttp compatibility
@@ -707,6 +710,7 @@ async def stream(
             json=json_body,
             data=request_data,
             timeout=timeout,
+            allow_redirects=self.follow_redirects,
         ) as response:
             # Add status_code property for backward compatibility
             if not hasattr(response, 'status_code'):