diff --git a/handlers/__init__.py b/handlers/__init__.py
index 237d6e6..7135bba 100644
--- a/handlers/__init__.py
+++ b/handlers/__init__.py
@@ -21,32 +21,50 @@ T = TypeVar("T", bound=Callable)
 BOT_MESSAGE_LENGTH = 4000
 
 
-def bot_reply_markdown(message: Message, hint: str, text: str, bot: TeleBot) -> None:
+def bot_reply_first(message: Message, who: str, bot: TeleBot) -> Message:
+    """Create the first reply message which make user feel the bot is working."""
+    return bot.reply_to(
+        message, f"*{who}* is _thinking_ \.\.\.", parse_mode="MarkdownV2"
+    )
+
+
+def bot_reply_markdown(reply_id: Message, who: str, text: str, bot: TeleBot) -> None:
     """
     reply the Markdown by take care of the message length.
     it will fallback to plain text in case of any failure
     """
     try:
         if len(text.encode("utf-8")) <= BOT_MESSAGE_LENGTH:
-            bot.reply_to(
-                message,
-                f"**{hint}**:\n{telegramify_markdown.convert(text)}",
+            bot.edit_message_text(
+                f"*{who}*:\n{telegramify_markdown.convert(text)}",
+                chat_id=reply_id.chat.id,
+                message_id=reply_id.message_id,
                 parse_mode="MarkdownV2",
             )
             return
 
         # Need a split of message
         msgs = smart_split(text, BOT_MESSAGE_LENGTH)
-        for i in range(len(msgs)):
+        bot.edit_message_text(
+            f"*{who}* \[1/{len(msgs)}\]:\n{telegramify_markdown.convert(msgs[0])}",
+            chat_id=reply_id.chat.id,
+            message_id=reply_id.message_id,
+            parse_mode="MarkdownV2",
+        )
+        for i in range(1, len(msgs)):
             bot.reply_to(
-                message,
-                f"**{hint}** \[{i+1}/{len(msgs)}\]:\n{telegramify_markdown.convert(msgs[i])}",
+                reply_id.reply_to_message,
+                f"*{who}* \[{i+1}/{len(msgs)}\]:\n{telegramify_markdown.convert(msgs[i])}",
                 parse_mode="MarkdownV2",
             )
     except Exception as e:
         print(traceback.format_exc())
         # print(f"wrong markdown format: {text}")
-        bot.reply_to(message, f"{hint}:\n{text}")
+        bot.edit_message_text(
+            f"*{who}*:\n{text}",
+            chat_id=reply_id.chat.id,
+            message_id=reply_id.message_id,
+        )
 
 
 def extract_prompt(message: str, bot_name: str) -> str:
@@ -133,4 +151,4 @@ def list_available_commands() -> list[str]:
 
 
 # `import *` will give you these
-__all__ = ["bot_reply_markdown", "extract_prompt"]
+__all__ = ["bot_reply_first", "bot_reply_markdown"]
diff --git a/handlers/claude.py b/handlers/claude.py
index 315caa6..ffae8e3 100644
--- a/handlers/claude.py
+++ b/handlers/claude.py
@@ -6,7 +6,7 @@ from anthropic import Anthropic, APITimeoutError
 from telebot import TeleBot
 from telebot.types import Message
 
-from . import bot_reply_markdown
+from . import *
 
 from telegramify_markdown import convert
 from telegramify_markdown.customize import markdown_symbol
@@ -37,6 +37,7 @@ def claude_handler(message: Message, bot: TeleBot) -> None:
         )
     else:
         player_message = claude_player_dict[str(message.from_user.id)]
+
     if m.strip() == "clear":
         bot.reply_to(
             message,
@@ -45,6 +46,9 @@ def claude_handler(message: Message, bot: TeleBot) -> None:
         player_message.clear()
         return
 
+    # show something, make it more responsible
+    reply_id = bot_reply_first(message, "Claude", bot)
+
     player_message.append({"role": "user", "content": m})
     # keep the last 5, every has two ask and answer.
     if len(player_message) > 10:
@@ -57,7 +61,7 @@ def claude_handler(message: Message, bot: TeleBot) -> None:
                 # tricky
                 player_message.pop()
         r = client.messages.create(
-            max_tokens=1024, messages=player_message, model=ANTHROPIC_MODEL
+            max_tokens=4096, messages=player_message, model=ANTHROPIC_MODEL
         )
         if not r.content:
             claude_reply_text = "Claude did not answer."
@@ -81,7 +85,7 @@ def claude_handler(message: Message, bot: TeleBot) -> None:
         player_message.clear()
         return
 
-    bot_reply_markdown(message, "Claude answer", claude_reply_text, bot)
+    bot_reply_markdown(reply_id, "Claude", claude_reply_text, bot)
 
 
 def claude_pro_handler(message: Message, bot: TeleBot) -> None:
@@ -94,7 +98,8 @@ def claude_pro_handler(message: Message, bot: TeleBot) -> None:
         )
     else:
         player_message = claude_pro_player_dict[str(message.from_user.id)]
-    if m.strip() == "clear":
+    q = m.strip()
+    if q == "clear" or len(q) == 0:
         bot.reply_to(
             message,
             "just clear you claude opus messages history",
@@ -113,7 +118,7 @@ def claude_pro_handler(message: Message, bot: TeleBot) -> None:
                 # tricky
                 player_message.pop()
         r = client.messages.create(
-            max_tokens=4096,
+            max_tokens=8192,
             messages=player_message,
             model=ANTHROPIC_PRO_MODEL,
             stream=True,
diff --git a/handlers/gemini.py b/handlers/gemini.py
index 84dc99c..ee1891c 100644
--- a/handlers/gemini.py
+++ b/handlers/gemini.py
@@ -6,7 +6,7 @@ from google.generativeai.types.generation_types import StopCandidateException
 from telebot import TeleBot
 from telebot.types import Message
 
-from . import bot_reply_markdown
+from . import *
 
 GOOGLE_GEMINI_KEY = environ.get("GOOGLE_GEMINI_KEY")
 
@@ -15,7 +15,7 @@ generation_config = {
     "temperature": 0.7,
     "top_p": 1,
     "top_k": 1,
-    "max_output_tokens": 4096,
+    "max_output_tokens": 8192,
 }
 
 safety_settings = [
@@ -56,6 +56,10 @@ def gemini_handler(message: Message, bot: TeleBot) -> None:
         )
         player.history.clear()
         return
+
+    # show something, make it more responsible
+    reply_id = bot_reply_first(message, "Gemini", bot)
+
     # keep the last 5, every has two ask and answer.
     if len(player.history) > 10:
         player.history = player.history[2:]
@@ -64,7 +68,8 @@ def gemini_handler(message: Message, bot: TeleBot) -> None:
         player.send_message(m)
         gemini_reply_text = player.last.text.strip()
         # Gemini is often using ':' in **Title** which not work in Telegram Markdown
-        gemini_reply_text = gemini_reply_text.replace(": **", "**\: ")
+        gemini_reply_text = gemini_reply_text.replace(":**", "\:**")
+        gemini_reply_text = gemini_reply_text.replace("：**", "**\: ")
     except StopCandidateException as e:
         match = re.search(r'content\s*{\s*parts\s*{\s*text:\s*"([^"]+)"', str(e))
         if match:
@@ -79,7 +84,7 @@ def gemini_handler(message: Message, bot: TeleBot) -> None:
             return
 
     # By default markdown
-    bot_reply_markdown(message, "Gemini answer", gemini_reply_text, bot)
+    bot_reply_markdown(reply_id, "Gemini", gemini_reply_text, bot)
 
 
 def gemini_photo_handler(message: Message, bot: TeleBot) -> None:
diff --git a/handlers/yi.py b/handlers/yi.py
index 6de7164..b22b570 100644
--- a/handlers/yi.py
+++ b/handlers/yi.py
@@ -7,7 +7,7 @@ import requests
 from telebot import TeleBot
 from telebot.types import Message
 
-from . import bot_reply_markdown
+from . import *
 
 YI_BASE_URL = environ.get("YI_BASE_URL")
 YI_API_KEY = environ.get("YI_API_KEY")
@@ -43,6 +43,9 @@ def yi_handler(message: Message, bot: TeleBot) -> None:
         player_message.clear()
         return
 
+    # show something, make it more responsible
+    reply_id = bot_reply_first(message, "Yi", bot)
+
     player_message.append({"role": "user", "content": m})
     # keep the last 5, every has two ask and answer.
     if len(player_message) > 10:
@@ -81,7 +84,7 @@ def yi_handler(message: Message, bot: TeleBot) -> None:
         return
 
     # reply back as Markdown and fallback to plain text if failed.
-    bot_reply_markdown(message, "yi answer", yi_reply_text, bot)
+    bot_reply_markdown(replay_id, "Yi", yi_reply_text, bot)
 
 
 def _image_to_data_uri(file_path):