diff --git a/litellm-rust/crates/cache-memory/src/cache.rs b/litellm-rust/crates/cache-memory/src/cache.rs
index 3b62c80b125..1908ff44a81 100644
--- a/litellm-rust/crates/cache-memory/src/cache.rs
+++ b/litellm-rust/crates/cache-memory/src/cache.rs
@@ -154,7 +154,7 @@ impl<V: Clone> InMemoryCache<V> {
         while let Some(Reverse((expiration, key))) = state.expiration_heap.peek().cloned() {
             if state.expirations.get(&key).copied() != Some(expiration) {
                 state.expiration_heap.pop();
-            } else if expiration < now {
+            } else if expiration <= now {
                 state.expiration_heap.pop();
                 Self::remove(state, &key);
             } else {
diff --git a/litellm-rust/crates/cache-memory/tests/cache.rs b/litellm-rust/crates/cache-memory/tests/cache.rs
index 423b8c959f2..aaf82641db7 100644
--- a/litellm-rust/crates/cache-memory/tests/cache.rs
+++ b/litellm-rust/crates/cache-memory/tests/cache.rs
@@ -45,6 +45,24 @@ fn default_explicit_and_override_ttls_follow_python_rules(clock: Arc<AtomicU64>)
     );
 }
 
+#[rstest]
+fn write_at_expiry_boundary_refreshes_ttl(clock: Arc<AtomicU64>) {
+    let cache = cache(clock.clone(), 4);
+    cache
+        .set_cache("key", "first".into(), Some(Duration::from_secs(10)))
+        .unwrap();
+    clock.store(110, Ordering::SeqCst);
+    cache
+        .set_cache("key", "second".into(), Some(Duration::from_secs(10)))
+        .unwrap();
+    assert_eq!(
+        cache.expires_at("key").unwrap(),
+        Some(Duration::from_secs(120))
+    );
+    clock.store(115, Ordering::SeqCst);
+    assert_eq!(cache.get_cache("key").unwrap(), Some("second".into()));
+}
+
 #[rstest]
 fn capacity_evicts_earliest_and_ignores_stale_heap_entries(clock: Arc<AtomicU64>) {
     let cache = cache(clock, 2);