From ce3fd8d84714e16c369acbfc60e86a45e4e6c890 Mon Sep 17 00:00:00 2001
From: bigsong <35025755+bigsongeth@users.noreply.github.com>
Date: Sat, 18 Jul 2026 15:43:39 +0800
Subject: [PATCH] =?UTF-8?q?feat(llm):=20=E6=96=B0=E5=A2=9E=20StepFun=20?=
 =?UTF-8?q?=E4=BE=9B=E5=BA=94=E5=95=86=E9=A2=84=E8=AE=BE=EF=BC=9B=E4=BF=AE?=
 =?UTF-8?q?=E5=A4=8D=20Codex=20OAuth=20=E9=BB=98=E8=AE=A4=E6=A8=A1?=
 =?UTF-8?q?=E5=9E=8B=E4=B8=8D=E5=8F=AF=E7=94=A8=20(#836)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

StepFun 预设：
- 供应商下拉新增 StepFun（阶跃星辰），预填 https://api.stepfun.com/v1，
  默认模型 step-1o-turbo-vision（无思考，流式首字 ~0.3s；step-3.x-flash
  系列为推理模型且思考无法关闭，正式内容需等隐藏思考结束，不适合听写润色）
- thinking 控制表新增 stepfun → reasoning_effort（开=medium / 关=low），
  base_url 兜底识别同步支持"自定义"preset 接入 StepFun 的场景
- 补默认 endpoint 映射与两个单元测试

Codex OAuth 默认模型：
- gpt-5.3-codex-spark 对 ChatGPT 账号会被后端 400 拒绝
  （"model is not supported when using Codex with a ChatGPT account"），
  导致每次润色静默失败回退原文；默认与占位改为实测可用的 gpt-5.5

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
---
 .../app/src-tauri/src/commands/credentials.rs |  1 +
 openless-all/app/src-tauri/src/polish.rs      | 53 ++++++++++++++++++-
 openless-all/app/src/i18n/en.ts               |  1 +
 openless-all/app/src/i18n/ja.ts               |  1 +
 openless-all/app/src/i18n/ko.ts               |  1 +
 openless-all/app/src/i18n/zh-CN.ts            |  1 +
 openless-all/app/src/i18n/zh-TW.ts            |  1 +
 .../src/pages/settings/ProvidersSection.tsx   | 17 +++++-
 8 files changed, 73 insertions(+), 3 deletions(-)

diff --git a/openless-all/app/src-tauri/src/commands/credentials.rs b/openless-all/app/src-tauri/src/commands/credentials.rs
index 305b0c91..7ed34ba1 100644
--- a/openless-all/app/src-tauri/src/commands/credentials.rs
+++ b/openless-all/app/src-tauri/src/commands/credentials.rs
@@ -92,6 +92,7 @@ fn llm_provider_default_endpoint(provider: &str) -> Option<&'static str> {
         "openrouterFree" => Some("https://openrouter.ai/api/v1"),
         "alibabaCoding" => Some("https://coding-intl.dashscope.aliyuncs.com/v1"),
         "codingPlanX" => Some("https://api.codingplanx.ai/v1"),
+        "stepfun" => Some("https://api.stepfun.com/v1"),
         _ => None,
     }
 }
diff --git a/openless-all/app/src-tauri/src/polish.rs b/openless-all/app/src-tauri/src/polish.rs
index 21c40d20..1ff47baf 100644
--- a/openless-all/app/src-tauri/src/polish.rs
+++ b/openless-all/app/src-tauri/src/polish.rs
@@ -25,7 +25,10 @@ const DEFAULT_REQUEST_TIMEOUT_SECS: u64 = 30;
 const BODY_PREVIEW_LIMIT: usize = 200;
 pub const CODEX_OAUTH_PROVIDER_ID: &str = "codex_oauth";
 pub const CODEX_DEFAULT_BASE_URL: &str = "https://chatgpt.com/backend-api";
-pub const CODEX_DEFAULT_MODEL: &str = "gpt-5.3-codex-spark";
+// 注意：gpt-5.3-codex-spark 不能做默认——ChatGPT 账号走 Codex OAuth 时后端会
+// 400 拒绝（"model is not supported when using Codex with a ChatGPT account"），
+// 每次润色都失败并回退原文。gpt-5.5 是该通道实测可用的模型。
+pub const CODEX_DEFAULT_MODEL: &str = "gpt-5.5";
 const CODEX_MIN_TOKEN_TTL_SECS: u64 = 60;
 
 #[derive(Clone, Debug)]
@@ -1607,7 +1610,9 @@ fn openai_compatible_thinking_control(provider_id: &str) -> Option<ThinkingContr
         "minimax" => Some(ThinkingControl::MiniMaxThinking),
         "openrouterFree" => Some(ThinkingControl::OpenRouterReasoning),
         "alibabaCoding" => Some(ThinkingControl::EnableThinking),
-        "openai" | "codingPlanX" => Some(ThinkingControl::ReasoningEffort),
+        // StepFun step-3.x-flash 系列按官方文档接受 reasoning_effort（low/medium/high，
+        // 无法完全关闭思考）；非推理模型（如 step-1o-turbo-vision）会忽略该字段。
+        "openai" | "codingPlanX" | "stepfun" => Some(ThinkingControl::ReasoningEffort),
         // custom / 其他未声明 provider 走 base_url 兜底识别——用户用自定义
         // endpoint 接入 MiniMax 时,根据 base_url 命中即下发官方 thinking 参数。
         _ => None,
@@ -1643,6 +1648,9 @@ fn openai_compatible_thinking_control_for_base_url(base_url: &str) -> Option<Thi
     if host.contains("dashscope") || host.contains("aliyuncs") {
         return Some(ThinkingControl::EnableThinking);
     }
+    if host.contains("stepfun") {
+        return Some(ThinkingControl::ReasoningEffort);
+    }
     None
 }
 
@@ -2650,6 +2658,47 @@ mod tests {
         }
     }
 
+    #[test]
+    fn openai_chat_body_adds_reasoning_effort_for_stepfun_channel() {
+        // StepFun 按渠道声明下发 reasoning_effort:开启思考发 medium,关闭发 low。
+        for (thinking_enabled, expected) in [(true, "medium"), (false, "low")] {
+            let provider = OpenAICompatibleLLMProvider::new(
+                OpenAICompatibleConfig::new(
+                    "stepfun",
+                    "StepFun",
+                    "https://api.stepfun.com/v1",
+                    "k",
+                    "step-3.7-flash",
+                )
+                .with_thinking_enabled(thinking_enabled),
+            );
+
+            let body = provider.chat_body(false, vec![json!({ "role": "user", "content": "hi" })]);
+
+            assert_eq!(body["reasoning_effort"], expected);
+        }
+    }
+
+    #[test]
+    fn openai_chat_body_falls_back_to_base_url_for_custom_stepfun_endpoint() {
+        // 用 "custom" preset + StepFun base_url 接入时,base_url 兜底识别需要
+        // 命中 "stepfun" 关键字,下发 reasoning_effort。
+        let provider = OpenAICompatibleLLMProvider::new(
+            OpenAICompatibleConfig::new(
+                "custom",
+                "Custom",
+                "https://api.stepfun.com/v1",
+                "k",
+                "step-3.7-flash",
+            )
+            .with_thinking_enabled(false),
+        );
+
+        let body = provider.chat_body(false, vec![json!({ "role": "user", "content": "hi" })]);
+
+        assert_eq!(body["reasoning_effort"], "low");
+    }
+
     #[test]
     fn openai_chat_body_omits_thinking_control_for_unknown_provider() {
         let provider = OpenAICompatibleLLMProvider::new(
diff --git a/openless-all/app/src/i18n/en.ts b/openless-all/app/src/i18n/en.ts
index 93cfbd36..1b130962 100644
--- a/openless-all/app/src/i18n/en.ts
+++ b/openless-all/app/src/i18n/en.ts
@@ -782,6 +782,7 @@ export const en: typeof zhCN = {
         alibabaCoding: 'Alibaba Cloud Coding Plan',
         codingPlanX: 'CodingPlanX',
         minimax: 'MiniMax (M3)',
+        stepfun: 'StepFun',
         custom: 'Custom',
         asrVolcengine: 'Volcengine bigasr',
         asrBailian: 'Alibaba Bailian realtime ASR',
diff --git a/openless-all/app/src/i18n/ja.ts b/openless-all/app/src/i18n/ja.ts
index 5e70c8dc..e3f156df 100644
--- a/openless-all/app/src/i18n/ja.ts
+++ b/openless-all/app/src/i18n/ja.ts
@@ -784,6 +784,7 @@ export const ja: typeof zhCN = {
         alibabaCoding: 'Alibaba Cloud Coding Plan',
         codingPlanX: 'CodingPlanX',
         minimax: 'MiniMax（M3）',
+        stepfun: 'StepFun（階躍星辰）',
         custom: 'カスタム',
         asrVolcengine: 'Volcengine bigasr',
         asrBailian: 'Alibaba Bailian リアルタイム ASR',
diff --git a/openless-all/app/src/i18n/ko.ts b/openless-all/app/src/i18n/ko.ts
index 832254d0..01c7fceb 100644
--- a/openless-all/app/src/i18n/ko.ts
+++ b/openless-all/app/src/i18n/ko.ts
@@ -784,6 +784,7 @@ export const ko: typeof zhCN = {
         alibabaCoding: 'Alibaba Cloud Coding Plan',
         codingPlanX: 'CodingPlanX',
         minimax: 'MiniMax (M3)',
+        stepfun: 'StepFun',
         custom: '사용자 정의',
         asrVolcengine: 'Volcengine bigasr',
         asrBailian: 'Alibaba Bailian 실시간 ASR',
diff --git a/openless-all/app/src/i18n/zh-CN.ts b/openless-all/app/src/i18n/zh-CN.ts
index 9202b600..f5ff528d 100644
--- a/openless-all/app/src/i18n/zh-CN.ts
+++ b/openless-all/app/src/i18n/zh-CN.ts
@@ -780,6 +780,7 @@ export const zhCN = {
         alibabaCoding: '阿里云 Coding Plan',
         codingPlanX: 'CodingPlanX',
         minimax: 'MiniMax（M3）',
+        stepfun: 'StepFun（阶跃星辰）',
         custom: '自定义',
         asrVolcengine: '火山引擎 bigasr',
         asrBailian: '阿里云百炼实时 ASR',
diff --git a/openless-all/app/src/i18n/zh-TW.ts b/openless-all/app/src/i18n/zh-TW.ts
index 222844ee..dde658f9 100644
--- a/openless-all/app/src/i18n/zh-TW.ts
+++ b/openless-all/app/src/i18n/zh-TW.ts
@@ -782,6 +782,7 @@ export const zhTW: typeof zhCN = {
         alibabaCoding: '阿里雲 Coding Plan',
         codingPlanX: 'CodingPlanX',
         minimax: 'MiniMax（M3）',
+        stepfun: 'StepFun（階躍星辰）',
         custom: '自定義',
         asrVolcengine: '火山引擎 bigasr',
         asrBailian: '阿里雲百煉即時 ASR',
diff --git a/openless-all/app/src/pages/settings/ProvidersSection.tsx b/openless-all/app/src/pages/settings/ProvidersSection.tsx
index db2cc303..c645346b 100644
--- a/openless-all/app/src/pages/settings/ProvidersSection.tsx
+++ b/openless-all/app/src/pages/settings/ProvidersSection.tsx
@@ -88,7 +88,9 @@ const LLM_PRESETS = [
     id: 'codex_oauth',
     nameKey: 'codexOAuth',
     baseUrl: '',
-    modelPlaceholder: 'gpt-5.3-codex-spark',
+    // gpt-5.3-codex-spark 对 ChatGPT 账号的 Codex 通道会被 400 拒绝，
+    // 默认与占位一律用实测可用的 gpt-5.5（见 polish.rs::CODEX_DEFAULT_MODEL）。
+    modelPlaceholder: 'gpt-5.5',
   },
   {
     id: 'mimo',
@@ -132,6 +134,19 @@ const LLM_PRESETS = [
     baseUrl: 'https://api.minimaxi.com/v1',
     modelPlaceholder: 'MiniMax-M3',
   },
+  {
+    // StepFun（阶跃星辰）OpenAI 兼容 /v1/chat/completions。
+    // 默认模型选 step-1o-turbo-vision：step-3.x-flash 系列是推理模型且思考无法关闭
+    // （reasoning_effort 只能调档，正式内容要等隐藏思考结束，润色场景 TTFT 2s+），
+    // 而 step-1o-turbo-vision 无思考、TTFT ~0.3s，润色忠实度实测更适合听写链路。
+    // provider_id 在后端 polish.rs::openai_compatible_thinking_control 命中
+    // "stepfun" → ReasoningEffort 分支；走"自定义"preset 接入时由 base_url
+    // 含 "stepfun" 兜底识别，见 polish.rs。
+    id: 'stepfun',
+    nameKey: 'stepfun',
+    baseUrl: 'https://api.stepfun.com/v1',
+    modelPlaceholder: 'step-1o-turbo-vision',
+  },
   {
     id: 'custom',
     nameKey: 'custom',