diff --git a/trustgraph-flow/trustgraph/model/text_completion/googleaistudio/llm.py b/trustgraph-flow/trustgraph/model/text_completion/googleaistudio/llm.py
index 1e9160ed..ca55a7e3 100644
--- a/trustgraph-flow/trustgraph/model/text_completion/googleaistudio/llm.py
+++ b/trustgraph-flow/trustgraph/model/text_completion/googleaistudio/llm.py
@@ -15,6 +15,7 @@ Input is prompt, output is response.
 from google import genai
 from google.genai import types
 from google.genai.types import HarmCategory, HarmBlockThreshold
+from google.genai.errors import ClientError
 from google.api_core.exceptions import ResourceExhausted
 import os
 import logging
@@ -152,6 +153,15 @@ class Processor(LlmService):
             # Leave rate limit retries to the default handler
             raise TooManyRequests()
 
+        except ClientError as e:
+            # google-genai SDK throws ClientError for 4xx errors
+            if e.code == 429:
+                logger.warning(f"Rate limit exceeded (ClientError 429): {e}")
+                raise TooManyRequests()
+            # Other client errors are unrecoverable
+            logger.error(f"GoogleAIStudio ClientError: {e}", exc_info=True)
+            raise e
+
         except Exception as e:
 
             # Apart from rate limits, treat all exceptions as unrecoverable
@@ -216,6 +226,15 @@ class Processor(LlmService):
             logger.warning("Rate limit exceeded during streaming")
             raise TooManyRequests()
 
+        except ClientError as e:
+            # google-genai SDK throws ClientError for 4xx errors
+            if e.code == 429:
+                logger.warning(f"Rate limit exceeded during streaming (ClientError 429): {e}")
+                raise TooManyRequests()
+            # Other client errors are unrecoverable
+            logger.error(f"GoogleAIStudio streaming ClientError: {e}", exc_info=True)
+            raise e
+
         except Exception as e:
             logger.error(f"GoogleAIStudio streaming exception ({type(e).__name__}): {e}", exc_info=True)
             raise e
diff --git a/trustgraph-vertexai/trustgraph/model/text_completion/vertexai/llm.py b/trustgraph-vertexai/trustgraph/model/text_completion/vertexai/llm.py
index 59aa5bfe..d7a7dd2a 100755
--- a/trustgraph-vertexai/trustgraph/model/text_completion/vertexai/llm.py
+++ b/trustgraph-vertexai/trustgraph/model/text_completion/vertexai/llm.py
@@ -16,6 +16,7 @@ import logging
 from google import genai
 from google.genai import types
 from google.genai.types import HarmCategory, HarmBlockThreshold
+from google.genai.errors import ClientError
 from google.api_core.exceptions import ResourceExhausted
 
 # Added for Anthropic model support
@@ -229,6 +230,15 @@ class Processor(LlmService):
             # Leave rate limit retries to the base handler
             raise TooManyRequests()
 
+        except ClientError as e:
+            # google-genai SDK throws ClientError for 4xx errors
+            if e.code == 429:
+                logger.warning(f"Hit rate limit (ClientError 429): {e}")
+                raise TooManyRequests()
+            # Other client errors are unrecoverable
+            logger.error(f"VertexAI ClientError: {e}", exc_info=True)
+            raise e
+
         except Exception as e:
             # Apart from rate limits, treat all exceptions as unrecoverable
             logger.error(f"VertexAI LLM exception: {e}", exc_info=True)
@@ -346,6 +356,15 @@ class Processor(LlmService):
             logger.warning(f"Hit rate limit during streaming: {e}")
             raise TooManyRequests()
 
+        except ClientError as e:
+            # google-genai SDK throws ClientError for 4xx errors
+            if e.code == 429:
+                logger.warning(f"Hit rate limit during streaming (ClientError 429): {e}")
+                raise TooManyRequests()
+            # Other client errors are unrecoverable
+            logger.error(f"VertexAI streaming ClientError: {e}", exc_info=True)
+            raise e
+
         except Exception as e:
             logger.error(f"VertexAI streaming exception: {e}", exc_info=True)
             raise e