diff --git a/balrog/client.py b/balrog/client.py
index ccabe97d..df387f48 100644
--- a/balrog/client.py
+++ b/balrog/client.py
@@ -147,9 +147,9 @@ def _initialize_client(self):
         if not self._initialized:
             if self.client_name.lower() == "vllm":
                 self.client = OpenAI(api_key="EMPTY", base_url=self.base_url)
-            elif self.client_name.lower() == "nvidia":
+            elif self.client_name.lower() == "nvidia" or self.client_name.lower() == "xai":
                 if not self.base_url or not self.base_url.strip():
-                    raise ValueError("base_url must be provided when using NVIDIA client")
+                    raise ValueError("base_url must be provided when using NVIDIA or XAI client")
                 self.client = OpenAI(base_url=self.base_url)
             elif self.client_name.lower() == "openai":
                 # For OpenAI, always use the standard API regardless of base_url
@@ -504,8 +504,8 @@ def create_llm_client(client_config):
 
     def client_factory():
         client_name_lower = client_config.client_name.lower()
-        if "openai" in client_name_lower or "vllm" in client_name_lower or "nvidia" in client_name_lower:
-            # NVIDIA uses OpenAI-compatible API, so we use the OpenAI wrapper
+        if "openai" in client_name_lower or "vllm" in client_name_lower or "nvidia" in client_name_lower or "xai" in client_name_lower:
+            # NVIDIA and XAI use OpenAI-compatible API, so we use the OpenAI wrapper
             return OpenAIWrapper(client_config)
         elif "gemini" in client_name_lower:
             return GoogleGenerativeAIWrapper(client_config)