router.py

@@ -418,7 +418,16 @@ async def enforce_router_api_key(request: Request, call_next):
         response.headers["Access-Control-Allow-Headers"] = "Authorization, Content-Type"
         response.headers["Access-Control-Allow-Methods"] = "GET, POST, PUT, DELETE, OPTIONS"
     return response
-        
+
+
+@app.exception_handler(openai.APIStatusError)
+async def _openai_api_status_error_handler(request: Request, exc: openai.APIStatusError):
+    """Forward upstream OpenAI-SDK status errors with their original status code and body
+    instead of letting them bubble up as 500s."""
+    body = exc.body if exc.body is not None else {"error": {"message": str(exc), "code": exc.status_code}}
+    return JSONResponse(status_code=exc.status_code, content=body)
+
+
 # -------------------------------------------------------------
 # 3. Global state: per‑endpoint per‑model active connection counters
 # -------------------------------------------------------------