diff --git a/common/chat.cpp b/common/chat.cpp
index 0cee80434e..8f42fef844 100644
--- a/common/chat.cpp
+++ b/common/chat.cpp
@@ -367,6 +367,13 @@ bool common_chat_templates_support_enable_thinking(const common_chat_templates *
     return params.supports_thinking;
 }
 
+// Check if the template source contains the preserve_thinking kwarg.
+// This is useful for printing a hint to the user to enable it.
+bool common_chat_templates_support_preserve_thinking(const common_chat_templates * chat_templates) {
+    std::string src = common_chat_templates_source(chat_templates);
+    return src.find("preserve_thinking") != std::string::npos;
+}
+
 std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const json & messages) {
     std::vector<common_chat_msg> msgs;
 
diff --git a/common/chat.h b/common/chat.h
index 7898f1623f..54010a45eb 100644
--- a/common/chat.h
+++ b/common/chat.h
@@ -347,6 +347,7 @@ common_reasoning_format common_reasoning_format_from_name(const std::string & fo
 common_chat_tool_choice common_chat_tool_choice_parse_oaicompat(const std::string & tool_choice);
 
 bool common_chat_templates_support_enable_thinking(const common_chat_templates * chat_templates);
+bool common_chat_templates_support_preserve_thinking(const common_chat_templates * chat_templates);
 
 // Parses a JSON array of messages in OpenAI's chat completion API format.
 std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const nlohmann::ordered_json & messages);
diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp
index 5c33a418f5..b731072b4b 100644
--- a/tools/server/server-context.cpp
+++ b/tools/server/server-context.cpp
@@ -1517,6 +1517,14 @@ private:
             const bool enable_thinking = params_base.enable_reasoning != 0 && template_supports_thinking;
             SRV_INF("%s: chat template, thinking = %d\n", __func__, enable_thinking);
 
+            // hint: suggest preserve_thinking if the template supports it but user hasn't set it
+            if (params_base.use_jinja && common_chat_templates_support_preserve_thinking(chat_templates.get())) {
+                auto it = params_base.default_template_kwargs.find("preserve_thinking");
+                if (it == params_base.default_template_kwargs.end()) {
+                    SRV_WRN("%s\n", "chat template supports 'preserve_thinking' - consider using --chat-template-kwargs \"{\\\"preserve_thinking\\\": true}\"");
+                }
+            }
+
             // IMPORTANT: chat_params is reused across sleeping / resuming states,
             //            never store llama_context/llama_model pointers in chat_params,
             //            as they may be invalidated after sleeping