diff --git a/common/chat.cpp b/common/chat.cpp
index 5b93c58873..1de8a3ba8f 100644
--- a/common/chat.cpp
+++ b/common/chat.cpp
@@ -1963,7 +1963,7 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
     params.add_generation_prompt = true;
     std::string gen_prompt       = common_chat_template_direct_apply_impl(tmpl, params);
     auto        diff             = calculate_diff_split(no_gen_prompt, gen_prompt);
-    params.generation_prompt     = diff.right;
+    params.generation_prompt     = diff.right + diff.suffix;
 
     params.add_generation_prompt = inputs.add_generation_prompt;
 
diff --git a/tests/test-chat.cpp b/tests/test-chat.cpp
index 72deeeab3c..814faa43ca 100644
--- a/tests/test-chat.cpp
+++ b/tests/test-chat.cpp
@@ -998,6 +998,7 @@ static void test_peg_parser(common_chat_templates *                      tmpls,
     auto parser = make_peg_parser(tmpls, tc.params, detailed_debug);
     if (detailed_debug) {
         LOG_DBG("Using parser: \n%s\n", parser.arena_.dump(parser.arena_.root()).c_str());
+        LOG_DBG("Generation prompt: '%s'\n", parser.params_.generation_prompt.c_str());
     }
 
     common_chat_msg msg_accum;
@@ -3102,8 +3103,19 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
     // Format: <minimax:tool_call><invoke name="func"><parameter name="key">value</parameter></invoke></minimax:tool_call>
     {
         auto tst = peg_tester("models/templates/MiniMax-M2.jinja", detailed_debug);
+        tst.test("</think>Hello, world!\nWhat's up?").enable_thinking(true).reasoning_format(COMMON_REASONING_FORMAT_AUTO).expect(message_assist).run();
+
+        tst.test("I'm\nthinking</think>Hello, world!\nWhat's up?").enable_thinking(true).reasoning_format(COMMON_REASONING_FORMAT_AUTO).expect(message_assist_thoughts).run();
+
+        tst.test("Let's call a tool:</think><minimax:tool_call>\n<invoke name=\"empty_args\">\n</invoke>\n</minimax:tool_call>").
+            enable_thinking(true).
+            reasoning_format(COMMON_REASONING_FORMAT_AUTO).
+            tools({ empty_args_tool }).
+            expect(message_with_reasoning_and_tool_call("Let's call a tool:", "empty_args", "{}")).
+            run();
+
         tst.test(
-               "<minimax:tool_call>\n<invoke name=\"special_function\">\n<parameter "
+               "</think><minimax:tool_call>\n<invoke name=\"special_function\">\n<parameter "
                "name=\"arg1\">1</parameter>\n</invoke>\n</minimax:tool_call>")
             .tools({ special_function_tool })
             .expect(message_assist_call)