diff --git a/fastdeploy/config.py b/fastdeploy/config.py
index b15a6dc824b..0c385cb6f2a 100644
--- a/fastdeploy/config.py
+++ b/fastdeploy/config.py
@@ -251,6 +251,24 @@ def __init__(
         # set attribute from pretrained_config
         for key, value in pretrained_config.items():
             setattr(self, key, value)
+
+        # Check NUM_MAX_DISPATCH_TOKENS_PER_RANK environment variable
+        env_num_max_dispatch = envs.NUM_MAX_DISPATCH_TOKENS_PER_RANK
+        if env_num_max_dispatch is not None:
+            # Only check consistency if model config explicitly sets this value
+            if "num_max_dispatch_tokens_per_rank" in pretrained_config:
+                model_num_max_dispatch = pretrained_config["num_max_dispatch_tokens_per_rank"]
+                if model_num_max_dispatch != env_num_max_dispatch:
+                    raise ValueError(
+                        f"num_max_dispatch_tokens_per_rank mismatch: "
+                        f"environment variable NUM_MAX_DISPATCH_TOKENS_PER_RANK={env_num_max_dispatch}, "
+                        f"but model config has num_max_dispatch_tokens_per_rank={model_num_max_dispatch}. "
+                        f"Please ensure they are consistent."
+                    )
+            else:
+                # Use env value if model config doesn't explicitly set it
+                self.num_max_dispatch_tokens_per_rank = env_num_max_dispatch
+
         # we need set default value when not exist
         for key, value in PRETRAINED_INIT_CONFIGURATION.items():
             if not hasattr(self, key):
diff --git a/fastdeploy/envs.py b/fastdeploy/envs.py
index 0c7ac3e22b1..9b71c165965 100644
--- a/fastdeploy/envs.py
+++ b/fastdeploy/envs.py
@@ -266,6 +266,13 @@ def _validate_split_kv_size(value: int) -> int:
     "FD_SAVE_OUTPUT_CACHE_FOR_PREEMPTED_REQUEST": lambda: bool(
         int(os.getenv("FD_SAVE_OUTPUT_CACHE_FOR_PREEMPTED_REQUEST", "1"))
     ),
+    # Number of max dispatch tokens per rank for MoE computation.
+    # If set, it must match the value in model config if present, otherwise an error will be raised.
+    "NUM_MAX_DISPATCH_TOKENS_PER_RANK": lambda: (
+        int(os.getenv("NUM_MAX_DISPATCH_TOKENS_PER_RANK", "0"))
+        if os.getenv("NUM_MAX_DISPATCH_TOKENS_PER_RANK")
+        else None
+    ),
 }