diff --git a/arraycontainer.go b/arraycontainer.go
index a5c28717..ae09a1a0 100644
--- a/arraycontainer.go
+++ b/arraycontainer.go
@@ -628,6 +628,30 @@ func (ac *arrayContainer) xor(a container) container {
 	panic("unsupported container type")
 }
 
+func (ac *arrayContainer) ixor(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.ixorArray(x)
+	case *bitmapContainer:
+		return ac.ixorBitmap(x)
+	case *runContainer16:
+		return ac.ixorRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) ixorArray(value2 *arrayContainer) container {
+	return ac.xorArray(value2)
+}
+
+func (ac *arrayContainer) ixorBitmap(value2 *bitmapContainer) container {
+	return value2.ixor(ac)
+}
+
+func (ac *arrayContainer) ixorRun16(value2 *runContainer16) container {
+	return value2.ixor(ac)
+}
+
 func (ac *arrayContainer) xorArray(value2 *arrayContainer) container {
 	value1 := ac
 	totalCardinality := value1.getCardinality() + value2.getCardinality()
diff --git a/benchmark_test.go b/benchmark_test.go
index d54d86e0..094cf1c6 100644
--- a/benchmark_test.go
+++ b/benchmark_test.go
@@ -1067,6 +1067,35 @@ func BenchmarkXorLopsided(b *testing.B) {
 	}
 }
 
+// BenchmarkXorDense benchmarks in-place Xor (ixor) on dense data
+// where containers are bitmapContainers (>4096 values per 16-bit chunk).
+// This is where in-place mutation of the bitmap[]uint64 slice should shine.
+func BenchmarkXorDense(b *testing.B) {
+	b.StopTimer()
+	r := rand.New(rand.NewSource(0))
+	// 50 chunks, each with ~10000 values → bitmapContainers (threshold is 4096)
+	numChunks := 50
+	valsPerChunk := 10000
+	s := NewBitmap()
+	for chunk := 0; chunk < numChunks; chunk++ {
+		base := uint32(chunk) * 65536
+		for i := 0; i < valsPerChunk; i++ {
+			s.Add(base + uint32(r.Intn(65536)))
+		}
+	}
+	x2 := NewBitmap()
+	for chunk := 0; chunk < numChunks; chunk++ {
+		base := uint32(chunk) * 65536
+		for i := 0; i < valsPerChunk; i++ {
+			x2.Add(base + uint32(r.Intn(65536)))
+		}
+	}
+	b.StartTimer()
+	for j := 0; j < b.N; j++ {
+		s.Xor(x2)
+	}
+}
+
 func BenchmarkBitmapReuseWithoutClear(b *testing.B) {
 	for j := 0; j < b.N; j++ {
 		s := NewBitmap()
diff --git a/bitmapcontainer.go b/bitmapcontainer.go
index 89682092..f4ea79f3 100644
--- a/bitmapcontainer.go
+++ b/bitmapcontainer.go
@@ -915,6 +915,43 @@ func (bc *bitmapContainer) iandBitmap(value2 *bitmapContainer) container {
 	return bc
 }
 
+func (bc *bitmapContainer) ixor(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.ixorArray(x)
+	case *bitmapContainer:
+		return bc.ixorBitmap(x)
+	case *runContainer16:
+		return bc.ixorRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) ixorArray(value2 *arrayContainer) container {
+	vbc := value2.toBitmapContainer()
+	return bc.ixorBitmap(vbc)
+}
+
+func (bc *bitmapContainer) ixorRun16(value2 *runContainer16) container {
+	rcb := value2.toBitmapContainer()
+	return bc.ixorBitmap(rcb)
+}
+
+func (bc *bitmapContainer) ixorBitmap(value2 *bitmapContainer) container {
+	newCardinality := int(popcntXorSlice(bc.bitmap, value2.bitmap))
+	if newCardinality > arrayDefaultMaxSize {
+		for k := 0; k < len(bc.bitmap); k++ {
+			bc.bitmap[k] = bc.bitmap[k] ^ value2.bitmap[k]
+		}
+		bc.cardinality = newCardinality
+		return bc
+	}
+	ac := newArrayContainerSize(newCardinality)
+	fillArrayXOR(ac.content, bc.bitmap, value2.bitmap)
+	ac.content = ac.content[:newCardinality]
+	return ac
+}
+
 func (bc *bitmapContainer) andNot(a container) container {
 	switch x := a.(type) {
 	case *arrayContainer:
diff --git a/roaring.go b/roaring.go
index f511539a..afb38bce 100644
--- a/roaring.go
+++ b/roaring.go
@@ -1509,8 +1509,7 @@ func (rb *Bitmap) Xor(x2 *Bitmap) {
 				pos1++
 				pos2++
 			} else {
-				// TODO: couple be computed in-place for reduced memory usage
-				c := rb.highlowcontainer.getContainerAtIndex(pos1).xor(x2.highlowcontainer.getContainerAtIndex(pos2))
+				c := rb.highlowcontainer.getWritableContainerAtIndex(pos1).ixor(x2.highlowcontainer.getContainerAtIndex(pos2))
 				if !c.isEmpty() {
 					rb.highlowcontainer.setContainerAtIndex(pos1, c)
 					pos1++
diff --git a/roaringarray.go b/roaringarray.go
index 31638d4b..49e1dd5c 100644
--- a/roaringarray.go
+++ b/roaringarray.go
@@ -39,6 +39,7 @@ type container interface {
 	not(start, final int) container        // range is [firstOfRange,lastOfRange)
 	inot(firstOfRange, endx int) container // i stands for inplace, range is [firstOfRange,endx)
 	xor(r container) container
+	ixor(r container) container // i stands for inplace
 	getShortIterator() shortPeekable
 	getUnsetIterator() shortPeekable
 	iterate(cb func(x uint16) bool) bool
diff --git a/runcontainer.go b/runcontainer.go
index a435c4d2..851c66dc 100644
--- a/runcontainer.go
+++ b/runcontainer.go
@@ -2416,6 +2416,30 @@ func (rc *runContainer16) xor(a container) container {
 	panic("unsupported container type")
 }
 
+func (rc *runContainer16) ixor(a container) container {
+	switch c := a.(type) {
+	case *arrayContainer:
+		return rc.ixorArray(c)
+	case *bitmapContainer:
+		return rc.ixorBitmap(c)
+	case *runContainer16:
+		return rc.ixorRunContainer16(c)
+	}
+	panic("unsupported container type")
+}
+
+func (rc *runContainer16) ixorArray(value2 *arrayContainer) container {
+	return rc.toBitmapContainer().ixor(value2)
+}
+
+func (rc *runContainer16) ixorBitmap(value2 *bitmapContainer) container {
+	return value2.ixor(rc)
+}
+
+func (rc *runContainer16) ixorRunContainer16(value2 *runContainer16) container {
+	return rc.toBitmapContainer().ixor(value2.toBitmapContainer())
+}
+
 func (rc *runContainer16) iandNot(a container) container {
 	switch c := a.(type) {
 	case *arrayContainer: