diff --git a/volc_limit_sync.py b/volc_limit_sync.py
index 709fbe1..f1b968d 100644
--- a/volc_limit_sync.py
+++ b/volc_limit_sync.py
@@ -827,16 +827,30 @@ class DailyLimitHandler(CustomLogger):
         if not data or not data.get("model"):
             return None
 
-        # 预留本次请求的预估消耗：优先真实 max_tokens（零采集损失），
-        # 未设置/非法时 fallback 常量，避免并发请求集体撞线超限。
+        # 预留本次请求的预估消耗 = 输入 token(精确) + 输出上限(max_tokens)
+        # 输入用 litellm.token_counter 精确计数(未知模型 fallback cl100k_base,
+        # 对中文偏高更安全);输出用请求声明的 max_tokens,未声明时 fallback 常量。
         try:
-            reserve = float(
+            max_out = float(
                 data.get("max_tokens")
                 or data.get("max_completion_tokens")
                 or PREBLOCK_RESERVE_TOKENS
             )
         except (TypeError, ValueError):
-            reserve = float(PREBLOCK_RESERVE_TOKENS)
+            max_out = float(PREBLOCK_RESERVE_TOKENS)
+
+        input_tokens = 0
+        try:
+            from litellm import token_counter as _tc
+            input_tokens = _tc(
+                model=data.get("model", ""),
+                messages=data.get("messages", []),
+            )
+        except Exception:
+            # token_counter 出错时保守:用 PREBLOCK_RESERVE_TOKENS 兜底输入
+            input_tokens = PREBLOCK_RESERVE_TOKENS
+
+        reserve = input_tokens + max_out
 
         blocked = _check_request_block(data["model"], reserve=reserve)
         if not blocked: