Merge a87ec75ba6 into fb3af2a08f

Removed deprecated Gemini models

app/api/common.ts

@@ -90,6 +90,14 @@ export async function requestOpenai(req: NextRequest) {
 
   const fetchUrl = cloudflareAIGatewayUrl(`${baseUrl}/${path}`);
   console.log("fetchUrl", fetchUrl);
+
+  let payload = await req.text();
+  if (baseUrl.includes("openrouter.ai")) {
+    const body = JSON.parse(payload);
+    body["include_reasoning"] = true;
+    payload = JSON.stringify(body);
+  }
+
   const fetchOptions: RequestInit = {
     headers: {
       "Content-Type": "application/json",
@@ -100,7 +108,7 @@ export async function requestOpenai(req: NextRequest) {
       }),
     },
     method: req.method,
-    body: req.body,
+    body: payload,
     // to fix #2485: https://stackoverflow.com/questions/55920957/cloudflare-worker-typeerror-one-time-use-body
     redirect: "manual",
     // @ts-ignore
@@ -111,10 +119,7 @@ export async function requestOpenai(req: NextRequest) {
   // #1815 try to refuse gpt4 request
   if (serverConfig.customModels && req.body) {
     try {
-      const clonedBody = await req.text();
-      fetchOptions.body = clonedBody;
-
-      const jsonBody = JSON.parse(clonedBody) as { model?: string };
+      const jsonBody = JSON.parse(payload) as { model?: string };
 
       // not undefined and is false
       if (

app/client/platforms/openai.ts

@@ -2,10 +2,10 @@
 // azure and openai, using same models. so using same LLMApi.
 import {
   ApiPath,
-  OPENAI_BASE_URL,
-  DEFAULT_MODELS,
-  OpenaiPath,
   Azure,
+  DEFAULT_MODELS,
+  OPENAI_BASE_URL,
+  OpenaiPath,
   REQUEST_TIMEOUT_MS,
   ServiceProvider,
 } from "@/app/constant";
@@ -18,13 +18,13 @@ import {
 } from "@/app/store";
 import { collectModelsWithDefaultModel } from "@/app/utils/model";
 import {
-  preProcessImageContent,
-  uploadImage,
   base64Image2Blob,
+  preProcessImageContent,
   streamWithThink,
+  uploadImage,
 } from "@/app/utils/chat";
 import { cloudflareAIGatewayUrl } from "@/app/utils/cloudflare";
-import { ModelSize, DalleQuality, DalleStyle } from "@/app/typing";
+import { DalleQuality, DalleStyle, ModelSize } from "@/app/typing";
 
 import {
   ChatOptions,
@@ -39,9 +39,9 @@ import Locale from "../../locales";
 import { getClientConfig } from "@/app/config/client";
 import {
   getMessageTextContent,
-  isVisionModel,
-  isDalle3 as _isDalle3,
   getTimeoutMSByModel,
+  isDalle3 as _isDalle3,
+  isVisionModel,
 } from "@/app/utils";
 import { fetch } from "@/app/utils/stream";
 
@@ -56,7 +56,7 @@ export interface OpenAIListModelResponse {
 
 export interface RequestPayload {
   messages: {
-    role: "system" | "user" | "assistant";
+    role: "developer" | "system" | "user" | "assistant";
     content: string | MultimodalContent[];
   }[];
   stream?: boolean;
@@ -238,8 +238,16 @@ export class ChatGPTApi implements LLMApi {
         // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
       };
 
-      // O1 使用 max_completion_tokens 控制token数 (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
       if (isO1OrO3) {
+        // by default the o1/o3 models will not attempt to produce output that includes markdown formatting
+        // manually add "Formatting re-enabled" developer message to encourage markdown inclusion in model responses
+        // (https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/reasoning?tabs=python-secure#markdown-output)
+        requestPayload["messages"].unshift({
+          role: "developer",
+          content: "Formatting re-enabled",
+        });
+
+        // o1/o3 uses max_completion_tokens to control the number of tokens (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
         requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
       }
 
@@ -295,6 +303,13 @@ export class ChatGPTApi implements LLMApi {
             useChatStore.getState().currentSession().mask?.plugin || [],
           );
         // console.log("getAsTools", tools, funcs);
+
+        // Add "include_reasoning" for OpenRouter: https://openrouter.ai/announcements/reasoning-tokens-for-thinking-models
+        if (chatPath.includes("openrouter.ai")) {
+          // @ts-ignore
+          requestPayload["include_reasoning"] = true;
+        }
+
         streamWithThink(
           chatPath,
           requestPayload,
@@ -311,6 +326,7 @@ export class ChatGPTApi implements LLMApi {
                 content: string;
                 tool_calls: ChatMessageTool[];
                 reasoning_content: string | null;
+                reasoning: string | null;
               };
             }>;
 
@@ -336,7 +352,9 @@ export class ChatGPTApi implements LLMApi {
               }
             }
 
-            const reasoning = choices[0]?.delta?.reasoning_content;
+            const reasoning =
+              choices[0]?.delta?.reasoning_content ||
+              choices[0]?.delta?.reasoning;
             const content = choices[0]?.delta?.content;
 
             // Skip if both content and reasoning_content are empty or null
@@ -412,6 +430,7 @@ export class ChatGPTApi implements LLMApi {
       options.onError?.(e as Error);
     }
   }
+
   async usage() {
     const formatDate = (d: Date) =>
       `${d.getFullYear()}-${(d.getMonth() + 1).toString().padStart(2, "0")}-${d
@@ -515,4 +534,5 @@ export class ChatGPTApi implements LLMApi {
     }));
   }
 }
+
 export { OpenaiPath };

app/constant.ts

@@ -523,20 +523,15 @@ const openaiModels = [
 ];
 
 const googleModels = [
-  "gemini-1.0-pro", // Deprecated on 2/15/2025
   "gemini-1.5-pro-latest",
   "gemini-1.5-pro",
   "gemini-1.5-pro-002",
-  "gemini-1.5-pro-exp-0827",
   "gemini-1.5-flash-latest",
   "gemini-1.5-flash-8b-latest",
   "gemini-1.5-flash",
   "gemini-1.5-flash-8b",
   "gemini-1.5-flash-002",
-  "gemini-1.5-flash-exp-0827",
   "learnlm-1.5-pro-experimental",
-  "gemini-exp-1114",
-  "gemini-exp-1121",
   "gemini-exp-1206",
   "gemini-2.0-flash",
   "gemini-2.0-flash-exp",
@@ -546,6 +541,7 @@ const googleModels = [
   "gemini-2.0-flash-thinking-exp-01-21",
   "gemini-2.0-pro-exp",
   "gemini-2.0-pro-exp-02-05",
+  "gemini-2.5-pro-preview-06-05",
 ];
 
 const anthropicModels = [
@@ -632,6 +628,18 @@ const xAIModes = [
   "grok-2-vision-1212",
   "grok-2-vision",
   "grok-2-vision-latest",
+  "grok-3-mini-fast-beta",
+  "grok-3-mini-fast",
+  "grok-3-mini-fast-latest",
+  "grok-3-mini-beta",
+  "grok-3-mini",
+  "grok-3-mini-latest",
+  "grok-3-fast-beta",
+  "grok-3-fast",
+  "grok-3-fast-latest",
+  "grok-3-beta",
+  "grok-3",
+  "grok-3-latest",
 ];
 
 const chatglmModels = [

app/utils/chat.ts

@@ -1,7 +1,7 @@
 import {
   CACHE_URL_PREFIX,
-  UPLOAD_URL,
   REQUEST_TIMEOUT_MS,
+  UPLOAD_URL,
 } from "@/app/constant";
 import { MultimodalContent, RequestMessage } from "@/app/client/api";
 import Locale from "@/app/locales";
@@ -111,6 +111,7 @@ export async function preProcessImageContentForAlibabaDashScope(
 }
 
 const imageCaches: Record<string, string> = {};
+
 export function cacheImageToBase64Image(imageUrl: string) {
   if (imageUrl.includes(CACHE_URL_PREFIX)) {
     if (!imageCaches[imageUrl]) {
@@ -385,6 +386,7 @@ export function stream(
       openWhenHidden: true,
     });
   }
+
   console.debug("[ChatAPI] start");
   chatApi(chatPath, headers, requestPayload, tools); // call fetchEventSource
 }
@@ -627,16 +629,9 @@ export function streamWithThink(
               if (remainText.length > 0) {
                 remainText += "\n";
               }
-              remainText += "> " + chunk.content;
-            } else {
-              // Handle newlines in thinking content
-              if (chunk.content.includes("\n\n")) {
-                const lines = chunk.content.split("\n\n");
-                remainText += lines.join("\n\n> ");
-              } else {
-                remainText += chunk.content;
-              }
+              remainText += "> ";
             }
+            remainText += chunk.content.replaceAll("\n", "\n> ");
           } else {
             // If in normal mode
             if (isInThinkingMode || isThinkingChanged) {
@@ -662,6 +657,7 @@ export function streamWithThink(
       openWhenHidden: true,
     });
   }
+
   console.debug("[ChatAPI] start");
   chatApi(chatPath, headers, requestPayload, tools); // call fetchEventSource
 }