feat(presets): real thinking switches for Qianfan and BytePlus Codex

2026-08-15 vendor audit, official-switch group:

- Baidu Qianfan Coding Plan: the v2 Chat API documents top-level
  thinking:{type:enabled/disabled} (same shape as Zhipu) across the
  plan's six main models, and Baidu's own OpenCode integration guide
  sends it on the /v2/coding endpoint. Declaration added (with the
  supportsThinking:true flag that gates the whole injection path) plus
  none/high levels. Known alias caveat noted inline: if the console
  resolves qianfan-code-latest to a non-thinking model the field is
  ignored, not rejected (platform documents unsupported params as
  ignore-not-error).
- BytePlus (ModelArk intl): the "not verified for the intl site"
  comment is obsolete - BytePlus' own Codex guide configures this exact
  coding/v3 endpoint with wire_api="responses". Switched apiFormat to
  openai_responses (aligned with the domestic Volcengine Plan presets,
  one less conversion layer) and pre-filled low/medium/high per the
  official Codex doc, with the glm-5-2 tier-collapse caveat noted.

Test moves BytePlus from the Chat-route list to the native Responses
list; reasoning-level expectations extended for both presets.

src/config/codexProviderPresets.ts

@@ -577,15 +577,20 @@ requires_openai_auth = true`,
     endpointCandidates: [
       "https://ark.ap-southeast.bytepluses.com/api/coding/v3",
     ],
-    // 国内站 plan/v3、coding/v3 已切原生 Responses（见 火山 Agent Plan /
-    // 火山 Coding Plan 预设），但 BytePlus
-    // 国际站（bytepluses.com）文档未单独核实，暂保持 Chat 路由
-    apiFormat: "openai_chat",
+    // 国际站已核实（2026-08-15 盘点）：BytePlus 官方 Codex 接入文档
+    //（docs.byteplus.com/en/docs/ModelArk/2556056）标准 config.toml 的
+    // base_url 就是本端点且 wire_api="responses"，OpenCode 文档亦明写
+    // Responses 优先——与国内站火山双 Plan 对齐切原生直连
+    apiFormat: "openai_responses",
     modelCatalog: modelCatalog([
       {
         model: "ark-code-latest",
         displayName: "Ark Code Latest",
         contextWindow: 256000,
+        // 官方 Codex 文档 model_reasoning_effort 限定 low/medium/high，与
+        // 国内站同名模型四份文档交叉印证。⚠️auto 路由别名固有不确定性：
+        // 路由到 glm-5-2-260617 时官方明载 low/medium 按 high 等价处理
+        reasoningLevels: ["low", "medium", "high"],
       },
     ]),
     category: "cn_official",
@@ -1137,12 +1142,26 @@ requires_openai_auth = true`,
     endpointCandidates: ["https://qianfan.baidubce.com/v2/coding"],
     apiFormat: "openai_chat",
     modelCatalog: modelCatalog([
+      // 两态（2026-08-15 盘点）：千帆 v2 官方 thinking:{type:enabled/disabled}
+      // 覆盖 Coding Plan 主力六模型，官方 OpenCode 接入文档在 /v2/coding 上
+      // 对 minimax-m2.5/glm-5/kimi-k2.5 照发该字段。⚠️别名固有缺陷：控制台把
+      // qianfan-code-latest 解析到 ernie-4.5-turbo 时 none 不会真关思考
       {
         model: "qianfan-code-latest",
         displayName: "Qianfan Code Latest",
         contextWindow: 131072,
+        reasoningLevels: ["none", "high"],
       },
     ]),
+    // 千帆 v2 Chat API 官方顶层参数（与智谱同形态）；平台对不支持的参数
+    // "忽略不报错"（官方多处明载），别名解析到非清单模型时只失效不 400
+    codexChatReasoning: {
+      supportsThinking: true,
+      supportsEffort: false,
+      thinkingParam: "thinking",
+      effortParam: "none",
+      outputFormat: "reasoning_content",
+    },
     category: "cn_official",
     icon: "baidu",
     iconColor: "#2932E1",

tests/config/codexChatProviderPresets.test.ts

@@ -10,15 +10,8 @@ const expectedChatPresets = new Map<
   string,
   { baseUrl: string; contextWindows: Record<string, number> }
 >([
-  // 火山 Agent Plan / Coding Plan（国内站 plan/v3、coding/v3）已切原生
-  // Responses，见下方 native 清单；BytePlus 国际站未核实，保持 Chat 路由
-  [
-    "BytePlus",
-    {
-      baseUrl: "https://ark.ap-southeast.bytepluses.com/api/coding/v3",
-      contextWindows: { "ark-code-latest": 256000 },
-    },
-  ],
+  // 火山 Agent Plan / Coding Plan 与 BytePlus 国际站（coding/v3）均已切
+  // 原生 Responses，见下方 native 清单
   [
     "Zhipu GLM",
     {
@@ -152,9 +145,11 @@ describe("Codex Chat provider presets", () => {
       { contextWindows: Record<string, number> }
     >([
       // 官方 Codex 文档确认 Agent Plan /api/plan/v3 与 Coding Plan
-      // /api/coding/v3 均支持 Responses API
+      // /api/coding/v3 均支持 Responses API；BytePlus 国际站 coding/v3
+      // 同（docs.byteplus.com/en/docs/ModelArk/2556056，2026-08-15 核实）
       ["火山 Agent Plan", { contextWindows: { "ark-code-latest": 256000 } }],
       ["火山 Coding Plan", { contextWindows: { "ark-code-latest": 256000 } }],
+      ["BytePlus", { contextWindows: { "ark-code-latest": 256000 } }],
       [
         "DouBaoSeed",
         { contextWindows: { "doubao-seed-2-1-pro-260628": 262144 } },

tests/config/codexReasoningLevelPresets.test.ts

@@ -72,6 +72,11 @@ describe("Codex preset pre-filled reasoning levels", () => {
     ["SiliconFlow en", "MiniMaxAI/MiniMax-M3", ["none", "high"]],
     // Novita：平台真开关 enable_thinking（声明已修正方言）→ 两态
     ["Novita AI", "zai-org/glm-5.1", ["none", "high"]],
+    // 千帆 v2 官方 thinking:{type}（声明已补）→ 两态
+    ["Baidu Qianfan Coding Plan", "qianfan-code-latest", ["none", "high"]],
+    // BytePlus 国际站已切原生 Responses，档位=官方 Codex 文档三档（与国内
+    // 站火山双 Plan 同源交叉印证）
+    ["BytePlus", "ark-code-latest", ["low", "medium", "high"]],
     // StepFun 官方两站模型页+reasoning 指南：3.7-flash 三档（默认 medium）、
     // 2603 两档；全系无关思考形态故无 none。effort 下发走后端 per-model
     // 推断（2603=low_high、3.7=passthrough）