Retry HTTP 400s that wrap a transient upstream code in error.code

OpenRouter is a passthrough for providers it does not control, and some transient
provider failures (502/503/529/...) arrive as HTTP 400 with the real status in the
JSON error.code. Retry only those; ordinary 400 invalid-request responses (including
inner error.code == 400) stay non-retryable. Reuses the existing TemporaryError/backoff.

Note: this edits generated code because retry classification here is HTTP-status only
(x-speakeasy-retries) and SDK hooks run outside the retry loop, so a 400 body cannot be
inspected via an overlay or hook. Preserving it across regen needs a genignore or upstream
Speakeasy support -- flagging for maintainers.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>

src/openrouter/utils/retries.py

@@ -1,6 +1,7 @@
 """Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT."""
 
 import asyncio
+import json
 import random
 import time
 from datetime import datetime
@@ -150,6 +151,24 @@ def _parse_retry_after_ms_header(response: httpx.Response) -> Optional[int]:
     return None
 
 
+TRANSIENT_INNER_ERROR_CODES = {408, 429, 500, 502, 503, 504, 524, 529}
+
+
+def _wraps_transient_inner_code(response: httpx.Response) -> bool:
+    """OpenRouter (a middleman) sometimes wraps a transient upstream failure in an HTTP
+    400 whose JSON error.code holds the real status; retry those. Ordinary 400
+    invalid-request responses (including inner error.code == 400) stay non-retryable."""
+    try:
+        body = response.json()
+    except (json.JSONDecodeError, httpx.ResponseNotRead):
+        return False
+    error = body.get("error") if isinstance(body, dict) else None
+    code = error.get("code") if isinstance(error, dict) else None
+    if isinstance(code, str) and code.isdigit():
+        code = int(code)
+    return code in TRANSIENT_INNER_ERROR_CODES
+
+
 def _get_sleep_interval(
     exception: Exception,
     initial_interval: int,
@@ -207,6 +226,12 @@ def retry(func, retries: Retries):
 
                         if res.status_code == parsed_code:
                             raise TemporaryError(res)
+
+                if res.status_code == 400:
+                    if not res.is_closed:
+                        res.read()
+                    if _wraps_transient_inner_code(res):
+                        raise TemporaryError(res)
             except (httpx.NetworkError, httpx.TimeoutException) as exception:
                 if retries.config.retry_connection_errors:
                     raise
@@ -252,6 +277,12 @@ async def retry_async(func, retries: Retries):
 
                         if res.status_code == parsed_code:
                             raise TemporaryError(res)
+
+                if res.status_code == 400:
+                    if not res.is_closed:
+                        await res.aread()
+                    if _wraps_transient_inner_code(res):
+                        raise TemporaryError(res)
             except (httpx.NetworkError, httpx.TimeoutException) as exception:
                 if retries.config.retry_connection_errors:
                     raise