diff --git a/src/opengradient/client/llm.py b/src/opengradient/client/llm.py
index 2d03512..e7e5337 100644
--- a/src/opengradient/client/llm.py
+++ b/src/opengradient/client/llm.py
@@ -1,5 +1,6 @@
 """LLM chat and completion via TEE-verified execution with x402 payments."""
 
+import base64
 import json
 import logging
 from dataclasses import dataclass
@@ -35,6 +36,7 @@
 _COMPLETION_ENDPOINT = "/v1/completions"
 _REQUEST_TIMEOUT = 60
 
+
 @dataclass(frozen=True)
 class _ChatParams:
     """Bundles the common parameters for chat/completion requests."""
@@ -385,8 +387,13 @@ async def _request() -> TextGenerationOutput:
                 headers=self._headers(params.x402_settlement_mode),
                 timeout=_REQUEST_TIMEOUT,
             )
-            response.raise_for_status()
             raw_body = await response.aread()
+            if response.status_code >= 400:
+                raise httpx.HTTPStatusError(
+                    _format_http_error(response, raw_body),
+                    request=response.request,
+                    response=response,
+                )
             result = json.loads(raw_body.decode())
 
             choices = result.get("choices")
@@ -532,3 +539,23 @@ async def _parse_sse_response(self, response, tee) -> AsyncGenerator[StreamChunk
                     chunk.tee_endpoint = tee.endpoint
                     chunk.tee_payment_address = tee.payment_address
                 yield chunk
+
+
+def _decode_payment_required(header_value: Optional[str]) -> str:
+    """Decode the base64-encoded JSON in the `payment-required` response header."""
+    if not header_value:
+        return "<missing>"
+    try:
+        decoded = base64.b64decode(header_value).decode("utf-8")
+        return json.dumps(json.loads(decoded), indent=2)
+    except Exception:
+        return header_value
+
+
+def _format_http_error(response: httpx.Response, body: bytes) -> str:
+    """Build an error message that surfaces the x402 payment-required details."""
+    return (
+        f"HTTP {response.status_code} from {response.url}\n"
+        f"Payment-Required: {_decode_payment_required(response.headers.get('payment-required'))}\n"
+        f"Body: {body.decode(errors='replace')}"
+    )
diff --git a/tests/llm_test.py b/tests/llm_test.py
index 8e5aba2..5309f28 100644
--- a/tests/llm_test.py
+++ b/tests/llm_test.py
@@ -7,7 +7,7 @@
 import json
 import ssl
 from contextlib import asynccontextmanager
-from typing import List
+from typing import Dict, List
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import httpx
@@ -80,6 +80,9 @@ class _FakeResponse:
     def __init__(self, status_code: int, body: bytes):
         self.status_code = status_code
         self._body = body
+        self.headers: Dict[str, str] = {}
+        self.request = MagicMock()
+        self.url = "https://test.tee.server/v1/chat/completions"
 
     def raise_for_status(self):
         pass