diff --git a/app/client/platforms/google.ts b/app/client/platforms/google.ts
index 53ff00aee..a7bce4fc2 100644
--- a/app/client/platforms/google.ts
+++ b/app/client/platforms/google.ts
@@ -29,7 +29,7 @@ import { RequestPayload } from "./openai";
 import { fetch } from "@/app/utils/stream";
 
 export class GeminiProApi implements LLMApi {
-  path(path: string): string {
+  path(path: string, shouldStream = false): string {
     const accessStore = useAccessStore.getState();
 
     let baseUrl = "";
@@ -51,8 +51,10 @@ export class GeminiProApi implements LLMApi {
     console.log("[Proxy Endpoint] ", baseUrl, path);
 
     let chatPath = [baseUrl, path].join("/");
+    if (shouldStream) {
+      chatPath += chatPath.includes("?") ? "&alt=sse" : "?alt=sse";
+    }
 
-    chatPath += chatPath.includes("?") ? "&alt=sse" : "?alt=sse";
     return chatPath;
   }
   extractMessage(res: any) {
@@ -60,6 +62,7 @@ export class GeminiProApi implements LLMApi {
 
     return (
       res?.candidates?.at(0)?.content?.parts.at(0)?.text ||
+      res?.at(0)?.candidates?.at(0)?.content?.parts.at(0)?.text ||
       res?.error?.message ||
       ""
     );
@@ -166,7 +169,10 @@ export class GeminiProApi implements LLMApi {
     options.onController?.(controller);
     try {
       // https://github.com/google-gemini/cookbook/blob/main/quickstarts/rest/Streaming_REST.ipynb
-      const chatPath = this.path(Google.ChatPath(modelConfig.model));
+      const chatPath = this.path(
+        Google.ChatPath(modelConfig.model),
+        shouldStream,
+      );
 
       const chatPayload = {
         method: "POST",
diff --git a/app/components/chat.tsx b/app/components/chat.tsx
index ed51d926f..51fe74fe7 100644
--- a/app/components/chat.tsx
+++ b/app/components/chat.tsx
@@ -960,9 +960,24 @@ function _Chat() {
           (scrollRef.current.scrollTop + scrollRef.current.clientHeight),
       ) <= 1
     : false;
+  const isAttachWithTop = useMemo(() => {
+    const lastMessage = scrollRef.current?.lastElementChild as HTMLElement;
+    // if scrolllRef is not ready or no message, return false
+    if (!scrollRef?.current || !lastMessage) return false;
+    const topDistance =
+      lastMessage!.getBoundingClientRect().top -
+      scrollRef.current.getBoundingClientRect().top;
+    // leave some space for user question
+    return topDistance < 100;
+  }, [scrollRef?.current?.scrollHeight]);
+
+  const isTyping = userInput !== "";
+
+  // if user is typing, should auto scroll to bottom
+  // if user is not typing, should auto scroll to bottom only if already at bottom
   const { setAutoScroll, scrollDomToBottom } = useScrollToBottom(
     scrollRef,
-    isScrolledToBottom,
+    (isScrolledToBottom || isAttachWithTop) && !isTyping,
   );
   const [hitBottom, setHitBottom] = useState(true);
   const isMobileScreen = useMobileScreen();
@@ -2071,6 +2086,6 @@ function _Chat() {
 
 export function Chat() {
   const chatStore = useChatStore();
-  const sessionIndex = chatStore.currentSessionIndex;
-  return <_Chat key={sessionIndex}></_Chat>;
+  const session = chatStore.currentSession();
+  return <_Chat key={session.id}></_Chat>;
 }
diff --git a/app/components/markdown.tsx b/app/components/markdown.tsx
index 8ea731123..ba85f0970 100644
--- a/app/components/markdown.tsx
+++ b/app/components/markdown.tsx
@@ -90,7 +90,11 @@ export function PreCode(props: { children: any }) {
     const refText = ref.current.querySelector("code")?.innerText;
     if (htmlDom) {
       setHtmlCode((htmlDom as HTMLElement).innerText);
-    } else if (refText?.startsWith("<!DOCTYPE")) {
+    } else if (
+      refText?.startsWith("<!DOCTYPE") ||
+      refText?.startsWith("<svg") ||
+      refText?.startsWith("<?xml")
+    ) {
       setHtmlCode(refText);
     }
   }, 600);
@@ -244,6 +248,10 @@ function escapeBrackets(text: string) {
 
 function tryWrapHtmlCode(text: string) {
   // try add wrap html code (fixed: html codeblock include 2 newline)
+  // ignore embed codeblock
+  if (text.includes("```")) {
+    return text;
+  }
   return text
     .replace(
       /([`]*?)(\w*?)([\n\r]*?)(<!DOCTYPE html>)/g,
diff --git a/app/utils.ts b/app/utils.ts
index 2dd80b8a3..1c359ef95 100644
--- a/app/utils.ts
+++ b/app/utils.ts
@@ -257,11 +257,11 @@ export function isVisionModel(model: string) {
   const excludeKeywords = ["claude-3-5-haiku-20241022"];
   const visionKeywords = [
     "vision",
-    "claude-3",
-    "gemini-1.5-pro",
-    "gemini-1.5-flash",
     "gpt-4o",
-    "gpt-4o-mini",
+    "claude-3",
+    "gemini-1.5",
+    "qwen-vl",
+    "qwen2-vl",
   ];
   const isGpt4Turbo =
     model.includes("gpt-4-turbo") && !model.includes("preview");