diff --git a/src/learn/learn.cpp b/src/learn/learn.cpp
index 7f18ff28..3942b606 100644
--- a/src/learn/learn.cpp
+++ b/src/learn/learn.cpp
@@ -93,6 +93,8 @@ namespace Learner
     static double elmo_lambda_high = 1.0;
     static double elmo_lambda_limit = 32000;
 
+    static double max_grad = 1.0;
+
     // Using stockfish's WDL with win rate model instead of sigmoid
     static bool use_wdl = false;
 
@@ -315,7 +317,7 @@ namespace Learner
             grad = lambda * (q - p) + (1.0 - lambda) * (q - t);
         }
 
-        return grad;
+        return std::clamp(grad, -max_grad, max_grad);
     }
 
     // Calculate cross entropy during learning
@@ -1072,6 +1074,7 @@ namespace Learner
             else if (option == "lambda") is >> elmo_lambda_low;
             else if (option == "lambda2") is >> elmo_lambda_high;
             else if (option == "lambda_limit") is >> elmo_lambda_limit;
+            else if (option == "max_grad") is >> max_grad;
 
             else if (option == "reduction_gameply") is >> params.reduction_gameply;
 
@@ -1175,6 +1178,7 @@ namespace Learner
         out << "  - elmo_lambda_low          : " << elmo_lambda_low << endl;
         out << "  - elmo_lambda_high         : " << elmo_lambda_high << endl;
         out << "  - elmo_lambda_limit        : " << elmo_lambda_limit << endl;
+        out << "  - max_grad                 : " << max_grad << endl;
         out << "  - eval_save_interval       : " << params.eval_save_interval << " sfens" << endl;
         out << "  - loss_output_interval     : " << params.loss_output_interval << " sfens" << endl;