diff --git a/litellm/router.py b/litellm/router.py index 3c2d31acd26..bd3170c70d7 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -10252,11 +10252,6 @@ class Router: return if api_base is None or "*" in model or params.get("use_clientside_credentials"): return - # Skip the /v1/models discovery call when the deployment already - # declares its token limits: discovery can only confirm values that - # are already configured, so the request is pure cost. On scale-to-zero - # backends (e.g. RunPod Serverless vLLM) each such call cold-starts a - # GPU worker and bills real money with no chat traffic (issue #42657). if self._deployment_token_limits_configured(raw_deployment.get("model_info")): return api_key: Final = params.api_key or dynamic_api_key