diff --git a/src/ai_scheduler.cpp b/src/ai_scheduler.cpp
index 54fa5c1..89591aa 100644
--- a/src/ai_scheduler.cpp
+++ b/src/ai_scheduler.cpp
@@ -99,6 +99,13 @@ ModelHandle AiScheduler::LoadModel(const std::string& model_path, std::string& e
         return kInvalidModelHandle;
     }
 
+    // Prefer using all NPU cores (RK3588).
+    // This does not change the shared-context behavior; it only hints RKNN runtime scheduling.
+    ret = rknn_set_core_mask(ctx->ctx, RKNN_NPU_CORE_0_1_2);
+    if (ret < 0) {
+        LogWarn("[AiScheduler] rknn_set_core_mask failed: " + std::to_string(ret));
+    }
+
     // Query input/output info
     rknn_input_output_num io_num;
     ret = rknn_query(ctx->ctx, RKNN_QUERY_IN_OUT_NUM, &io_num, sizeof(io_num));