diff --git a/modelscope/hub/api.py b/modelscope/hub/api.py
index 2de69a44..8a7b8409 100644
--- a/modelscope/hub/api.py
+++ b/modelscope/hub/api.py
@@ -32,13 +32,16 @@ from requests.adapters import HTTPAdapter, Retry
 from requests.exceptions import HTTPError
 from tqdm.auto import tqdm
 
-from modelscope.hub.constants import (API_HTTP_CLIENT_MAX_RETRIES,
+from modelscope.hub.constants import (API_HTTP_CLIENT_CONNECT_TIMEOUT,
+                                      API_HTTP_CLIENT_MAX_RETRIES,
                                       API_HTTP_CLIENT_TIMEOUT,
                                       API_RESPONSE_FIELD_DATA,
                                       API_RESPONSE_FIELD_EMAIL,
                                       API_RESPONSE_FIELD_GIT_ACCESS_TOKEN,
                                       API_RESPONSE_FIELD_MESSAGE,
                                       API_RESPONSE_FIELD_USERNAME,
+                                      CREATE_TAG_MAX_RETRIES,
+                                      CREATE_TAG_RETRY_BACKOFF,
                                       DEFAULT_MAX_WORKERS,
                                       DEFAULT_MODELSCOPE_INTL_DOMAIN,
                                       MODELSCOPE_CLOUD_ENVIRONMENT,
@@ -497,11 +500,46 @@ class HubApi:
                 'Ref': revision
             }
 
-        r = self.session.post(
-            path,
-            json=body,
-            cookies=cookies,
-            headers=self.builder_headers(self.headers))
+        tag_timeout = (API_HTTP_CLIENT_CONNECT_TIMEOUT,
+                       API_HTTP_CLIENT_TIMEOUT)
+
+        retryable_status = {500, 502, 503, 504}
+        attempts = max(1, CREATE_TAG_MAX_RETRIES)
+        r = None
+        for attempt in range(1, attempts + 1):
+            retry_reason = None
+            try:
+                r = self.session.post(
+                    path,
+                    json=body,
+                    cookies=cookies,
+                    headers=self.builder_headers(self.headers),
+                    timeout=tag_timeout)
+            except (requests.exceptions.ReadTimeout,
+                    requests.exceptions.ConnectTimeout,
+                    requests.exceptions.ConnectionError) as e:
+                if attempt >= attempts:
+                    logger.error(
+                        f'create_model_tag POST failed after {attempts} '
+                        f'attempt(s) due to transient network error: {e}. '
+                        f'Consider raising MODELSCOPE_API_HTTP_CLIENT_TIMEOUT '
+                        f'(current={API_HTTP_CLIENT_TIMEOUT}s) or '
+                        f'MODELSCOPE_CREATE_TAG_MAX_RETRIES '
+                        f'(current={CREATE_TAG_MAX_RETRIES}).')
+                    raise
+                retry_reason = f'{type(e).__name__}: {e}'
+            else:
+                if r.status_code in retryable_status and attempt < attempts:
+                    retry_reason = (f'retryable HTTP {r.status_code} '
+                                    f'from server')
+                else:
+                    break
+
+            sleep_s = CREATE_TAG_RETRY_BACKOFF * (2 ** (attempt - 1))
+            logger.warning(
+                f'create_model_tag POST attempt {attempt}/{attempts} '
+                f'failed with {retry_reason}. Retrying in {sleep_s}s...')
+            time.sleep(sleep_s)
 
         raise_for_http_status(r)
         d = r.json()
diff --git a/modelscope/hub/constants.py b/modelscope/hub/constants.py
index ae6a5d72..85ad69d4 100644
--- a/modelscope/hub/constants.py
+++ b/modelscope/hub/constants.py
@@ -19,10 +19,23 @@ DEFAULT_CREDENTIALS_PATH = Path.home().joinpath('.modelscope', 'credentials')
 MODELSCOPE_CREDENTIALS_PATH = os.environ.get(
     'MODELSCOPE_CREDENTIALS_PATH', DEFAULT_CREDENTIALS_PATH.as_posix())
 REQUESTS_API_HTTP_METHOD = ['get', 'head', 'post', 'put', 'patch', 'delete']
-API_HTTP_CLIENT_TIMEOUT = 60
+# Default per-socket timeout (seconds) applied to all session HTTP methods
+# in HubApi.__init__. User-tunable via env to mitigate transient ReadTimeout
+# on heavy server-side ops (e.g. create_model_tag which performs git ops on
+# the remote repo). Default raised to 90 to mitigate transient timeouts.
+API_HTTP_CLIENT_TIMEOUT = int(
+    os.environ.get('MODELSCOPE_API_HTTP_CLIENT_TIMEOUT', 90))
+
+API_HTTP_CLIENT_CONNECT_TIMEOUT = int(
+    os.environ.get('MODELSCOPE_API_HTTP_CLIENT_CONNECT_TIMEOUT', 10))
 API_HTTP_CLIENT_MAX_RETRIES = int(
     os.environ.get('API_HTTP_CLIENT_MAX_RETRIES', 5))
 
+CREATE_TAG_MAX_RETRIES = int(
+    os.environ.get('MODELSCOPE_CREATE_TAG_MAX_RETRIES', 3))
+CREATE_TAG_RETRY_BACKOFF = int(
+    os.environ.get('MODELSCOPE_CREATE_TAG_RETRY_BACKOFF', 2))
+
 # Application-level retry for blob upload
 UPLOAD_BLOB_MAX_RETRIES = int(os.environ.get('UPLOAD_BLOB_MAX_RETRIES', 5))
 UPLOAD_BLOB_RETRY_BACKOFF = int(os.environ.get('UPLOAD_BLOB_RETRY_BACKOFF', 2))