From 106db97f8c5be534c29dc8bde7bdfdfa738c45b7 Mon Sep 17 00:00:00 2001
From: Davidlasky <davidjiangxc@gmail.com>
Date: Tue, 1 Apr 2025 17:47:16 -0500
Subject: [PATCH] enable o3-mini-high, optimize timeout for thinking models

---
 app/client/platforms/openai.ts | 15 +++++++--------
 app/utils.ts                   | 10 +++++++++-
 2 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/app/client/platforms/openai.ts b/app/client/platforms/openai.ts
index b6c2393e6..72e882467 100644
--- a/app/client/platforms/openai.ts
+++ b/app/client/platforms/openai.ts
@@ -241,18 +241,18 @@ export class ChatGPTApi implements LLMApi {
 
       // O1 使用 max_completion_tokens 控制token数 (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
       if (isO1OrO3) {
-        requestPayload["max_completion_tokens"] = 25000;
+        requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
       }
 
       if (isO3) {
         requestPayload["reasoning_effort"] = "high";
+        // make o3-mini defaults to high reasoning effort
       }
 
       // add max_tokens to vision model
       if (visionModel) {
         if (isO1) {
           requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
-
         } else {
           requestPayload["max_tokens"] = Math.max(modelConfig.max_tokens, 4000);
         }
@@ -297,6 +297,11 @@ export class ChatGPTApi implements LLMApi {
           isDalle3 ? OpenaiPath.ImagePath : OpenaiPath.ChatPath,
         );
       }
+      // make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        getTimeoutMSByModel(options.config.model),
+      );
       if (shouldStream) {
         let index = -1;
         const [tools, funcs] = usePluginStore
@@ -404,12 +409,6 @@ export class ChatGPTApi implements LLMApi {
           headers: getHeaders(),
         };
 
-        // make a fetch request
-        const requestTimeoutId = setTimeout(
-          () => controller.abort(),
-          getTimeoutMSByModel(options.config.model),
-        );
-
         const res = await fetch(chatPath, chatPayload);
         clearTimeout(requestTimeoutId);
 
diff --git a/app/utils.ts b/app/utils.ts
index fed0c953e..f2454e5cc 100644
--- a/app/utils.ts
+++ b/app/utils.ts
@@ -306,8 +306,16 @@ export function getTimeoutMSByModel(model: string) {
     model.includes("deepseek-r") ||
     model.includes("-thinking") ||
     model.includes("pro")
-  )
+  ) {
+    console.log(
+      "thinking model is " +
+        model +
+        " timeout is " +
+        REQUEST_TIMEOUT_MS_FOR_THINKING,
+    );
     return REQUEST_TIMEOUT_MS_FOR_THINKING;
+  }
+  console.log("normal model is " + model + " timeout is " + REQUEST_TIMEOUT_MS);
   return REQUEST_TIMEOUT_MS;
 }