diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2026-07-14-20-19-48.gh-issue-152075.V9Rwhp.rst b/Misc/NEWS.d/next/Core_and_Builtins/2026-07-14-20-19-48.gh-issue-152075.V9Rwhp.rst
new file mode 100644
index 00000000000000..84c7d36b97acba
--- /dev/null
+++ b/Misc/NEWS.d/next/Core_and_Builtins/2026-07-14-20-19-48.gh-issue-152075.V9Rwhp.rst
@@ -0,0 +1,2 @@
+Reduced lock contention during LOAD_GLOBAL bytecode specialization under
+free threading.
diff --git a/Python/specialize.c b/Python/specialize.c
index 773b55c329c772..05cb76ff015ff4 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -1457,6 +1457,14 @@ _Py_Specialize_LoadGlobal(
     PyObject *globals, PyObject *builtins,
     _Py_CODEUNIT *instr, PyObject *name)
 {
+#ifdef Py_GIL_DISABLED
+    if (PyMutex_IsLocked(&globals->ob_mutex) || PyMutex_IsLocked(&builtins->ob_mutex)) {
+        // Skip specialization if either dictionary is locked to avoid lock
+        // contention.
+        unspecialize(instr);
+        return;
+    }
+#endif
     Py_BEGIN_CRITICAL_SECTION2(globals, builtins);
     specialize_load_global_lock_held(globals, builtins, instr, name);
     Py_END_CRITICAL_SECTION2();