diff --git a/litellm/llms/zai/responses/transformation.py b/litellm/llms/zai/responses/transformation.py
index fc562bd2cb3..c47b8b5943a 100644
--- a/litellm/llms/zai/responses/transformation.py
+++ b/litellm/llms/zai/responses/transformation.py
@@ -20,6 +20,8 @@ class ZAIResponsesAPIConfig(OpenAIResponsesAPIConfig):
     ``Authorization: Bearer`` header.
     """
 
+    _ZAI_CHAT_API_BASE_SUFFIXES: Final = ("/api/paas/v4", "/api/coding/paas/v4")
+
     @property
     def custom_llm_provider(self) -> LlmProviders:
         return LlmProviders.ZAI
@@ -44,7 +46,17 @@ class ZAIResponsesAPIConfig(OpenAIResponsesAPIConfig):
         api_base: str | None,
         litellm_params: dict,
     ) -> str:
-        base_url = api_base or get_secret_str("ZAI_RESPONSES_API_BASE") or "https://api.z.ai/api/v1"
+        # ``litellm_params.api_base`` can carry the Z.AI chat-completions base
+        # (``/api/paas/v4``) when the generic provider resolver pre-fills it from
+        # the chat config. Z.AI serves Responses on a different base, so ignore
+        # the chat-only bases and use the Responses base instead.
+        normalized_api_base = (api_base or "").rstrip("/")
+        chat_base_passed_in: Final = normalized_api_base.endswith(self._ZAI_CHAT_API_BASE_SUFFIXES)
+        base_url = (
+            api_base
+            if api_base and not chat_base_passed_in
+            else get_secret_str("ZAI_RESPONSES_API_BASE") or "https://api.z.ai/api/v1"
+        )
 
         base_url = base_url.rstrip("/")
         if base_url.endswith("/responses"):
diff --git a/tests/test_litellm/llms/zai/test_zai_responses_transformation.py b/tests/test_litellm/llms/zai/test_zai_responses_transformation.py
index 0925ff21a22..064bc21d9b5 100644
--- a/tests/test_litellm/llms/zai/test_zai_responses_transformation.py
+++ b/tests/test_litellm/llms/zai/test_zai_responses_transformation.py
@@ -30,6 +30,46 @@ def test_zai_responses_url_defaults_to_responses_endpoint(monkeypatch):
         assert config.get_complete_url(api_base=api_base, litellm_params={}) == expected_url
 
 
+def test_zai_responses_url_ignores_chat_completions_api_base(monkeypatch):
+    monkeypatch.delenv("ZAI_RESPONSES_API_BASE", raising=False)
+    config = ZAIResponsesAPIConfig()
+
+    chat_bases = (
+        "https://api.z.ai/api/paas/v4",
+        "https://api.z.ai/api/paas/v4/",
+        "https://api.z.ai/api/coding/paas/v4",
+    )
+
+    for chat_base in chat_bases:
+        assert config.get_complete_url(api_base=chat_base, litellm_params={}) == "https://api.z.ai/api/v1/responses"
+
+
+def test_zai_responses_url_keeps_custom_api_base(monkeypatch):
+    monkeypatch.delenv("ZAI_RESPONSES_API_BASE", raising=False)
+    config = ZAIResponsesAPIConfig()
+
+    assert (
+        config.get_complete_url(
+            api_base="https://gateway.example.com/openai/v1",
+            litellm_params={},
+        )
+        == "https://gateway.example.com/openai/v1/responses"
+    )
+
+
+def test_zai_responses_url_env_overrides_chat_completions_api_base(monkeypatch):
+    monkeypatch.setenv("ZAI_RESPONSES_API_BASE", "https://gateway.example.com/responses-root")
+    config = ZAIResponsesAPIConfig()
+
+    assert (
+        config.get_complete_url(
+            api_base="https://api.z.ai/api/paas/v4",
+            litellm_params={},
+        )
+        == "https://gateway.example.com/responses-root/responses"
+    )
+
+
 def test_zai_responses_headers_use_bearer_token():
     config = ZAIResponsesAPIConfig()
     litellm_params = GenericLiteLLMParams(api_key="sk-zai")