diff --git a/keras/src/losses/loss.py b/keras/src/losses/loss.py
index 227c43b2128..e0197a6bb8a 100644
--- a/keras/src/losses/loss.py
+++ b/keras/src/losses/loss.py
@@ -123,7 +123,7 @@ def squeeze_or_expand_to_same_rank(x1, x2, expand_rank_1=True):
     return x1, x2
 
 
-def reduce_values(values, reduction="sum_over_batch_size"):
+def reduce_values(values, reduction="sum_over_batch_size", sample_weight=None):
     if (
         reduction is None
         or reduction == "none"
@@ -132,7 +132,12 @@ def reduce_values(values, reduction="sum_over_batch_size"):
     ):
         return values
     loss = ops.sum(values)
-    if reduction == "sum_over_batch_size":
+    if sample_weight is not None and reduction == "sum_over_batch_size":
+        loss /= ops.cast(
+            ops.sum(sample_weight),
+            loss.dtype,
+        )
+    elif reduction == "sum_over_batch_size":
         loss /= ops.cast(
             ops.prod(ops.convert_to_tensor(ops.shape(values), dtype="int32")),
             loss.dtype,
@@ -169,7 +174,7 @@ def reduce_weighted_values(
         values = values * sample_weight
 
     # Apply reduction function to the individual weighted losses.
-    loss = reduce_values(values, reduction)
+    loss = reduce_values(values, reduction, sample_weight)
     return loss