diff --git a/vllm_omni/entrypoints/cli/main.py b/vllm_omni/entrypoints/cli/main.py
index 79129d732..17be480b2 100644
--- a/vllm_omni/entrypoints/cli/main.py
+++ b/vllm_omni/entrypoints/cli/main.py
@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 """
 CLI entry point for vLLM-Omni that intercepts vLLM commands.
 """
@@ -24,6 +27,7 @@ def main():
             os.environ["VLLM_LOGGING_COLOR"] = "1"
 
         from vllm.entrypoints.serve.utils.api_utils import VLLM_SUBCMD_PARSER_EPILOG, cli_env_setup
+        from vllm.logger import configure_logging_from_args
 
         import vllm_omni.entrypoints.cli.benchmark.main
         import vllm_omni.entrypoints.cli.serve
@@ -66,6 +70,7 @@ def main():
                 cmds[cmd.name] = cmd
         args = parser.parse_args()
         if args.subparser in cmds:
+            configure_logging_from_args(args)
             cmds[args.subparser].validate(args)
 
         if hasattr(args, "dispatch_function"):
diff --git a/vllm_omni/entrypoints/openai/api_server.py b/vllm_omni/entrypoints/openai/api_server.py
index 88bcac8cb..189323b66 100644
--- a/vllm_omni/entrypoints/openai/api_server.py
+++ b/vllm_omni/entrypoints/openai/api_server.py
@@ -80,7 +80,7 @@ from vllm.entrypoints.speech_to_text.transcription.serving import (
 from vllm.entrypoints.speech_to_text.translation.serving import (
     OpenAIServingTranslation,
 )
-from vllm.logger import init_logger
+from vllm.logger import configure_logging_from_args, init_logger
 from vllm.renderers.online_renderer import OnlineRenderer
 from vllm.tasks import POOLING_TASKS
 from vllm.tool_parsers import ToolParserManager
@@ -3109,6 +3109,7 @@ if __name__ == "__main__":
     # when __main__ is called, i.e., --omni is only used when called through the entrypoints.
     parser.add_argument("--omni", action="store_true", default=False)
     args = parser.parse_args()
+    configure_logging_from_args(args)
     # sync args.model to model_tag, because if we pass the model positionally,
     # args.model will be the default from vLLM's ModelConfig (currently
     # Qwen/Qwen3-0.6B) and crash cryptically.