luckytri 1.0.4 → 1.0.6

package/server/core/model-presets.js

@@ -1,900 +1,900 @@
-// Checked against vendor API docs on 2026-09-23.
-// maxOutputTokens is the published output ceiling and maxInputTokens is what
-// the window leaves after it, so every preset passes validateModel as is.
-
-function limits(contextWindow, maxOutputTokens) {
-  return {
-    contextWindow,
-    maxOutputTokens,
-    maxInputTokens: contextWindow - maxOutputTokens,
-  };
-}
-
-// OpenAI and Bedrock bill the whole GPT request at the long-context rate once
-// input passes 272K tokens. The standard tier stops input there and keeps the
-// full 128K output.
-function gptWindows() {
-  const standard = {
-    contextWindow: 272000 + 128000,
-    maxInputTokens: 272000,
-    maxOutputTokens: 128000,
-  };
-  return {
-    ...standard,
-    contextWindows: [
-      { label: "标准（输入 ≤272K）", ...standard },
-      { label: "百万（1.05M）", ...limits(1050000, 128000) },
-    ],
-  };
-}
-
-const shared = {
-  embedding: false,
-  embeddingModel: "",
-  timeoutMs: 120000,
-  system: true,
-  json: true,
-  tools: true,
-};
-
-const OPENAI = "https://api.openai.com/v1";
-const BEDROCK = "https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1";
-const OPENROUTER = "https://openrouter.ai/api/v1";
-const OPENCODE_GO = "https://opencode.ai/zen/go/v1";
-const OPENCODE_ZEN = "https://opencode.ai/zen/v1";
-const VOLCENGINE_CODING_PLAN =
-  "https://ark.cn-beijing.volces.com/api/coding/v3";
-const GPT_EFFORTS = ["none", "low", "medium", "high", "xhigh", "max"];
-const ASTRA_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
-
-function gpt(fields) {
-  return {
-    ...gptWindows(),
-    ...shared,
-    vision: true,
-    thinkingStyle: "openai",
-    tokenField: "max_completion_tokens",
-    reasoningEfforts: GPT_EFFORTS,
-    reasoningEffort: "medium",
-    temperature: 1,
-    topP: 1,
-    ...fields,
-  };
-}
-
-function openRouterGpt(fields) {
-  const profile = gpt({
-    ...fields,
-    provider: "openrouter",
-    baseUrl: OPENROUTER,
-    contextWindow: 1050000,
-    maxInputTokens: 922000,
-    maxOutputTokens: 128000,
-  });
-  delete profile.contextWindows;
-  return profile;
-}
-
-function openCodeModel({
-  channel,
-  model,
-  label,
-  protocol = "chat",
-  summary,
-  vision = false,
-  reasoningEfforts,
-  reasoningEffort = "none",
-  limits: modelLimits = limits(128000, 8192),
-  temperature = 0.85,
-  topP = 1,
-}) {
-  const isGo = channel === "go";
-  const anthropic = protocol === "anthropic";
-  return {
-    id: `opencode-${channel}-${model}`,
-    vendor: isGo ? "OpenCode Go" : "OpenCode Zen",
-    label,
-    summary,
-    provider: isGo ? "opencode-go" : "opencode-zen",
-    baseUrl: isGo ? OPENCODE_GO : OPENCODE_ZEN,
-    model,
-    apiProtocol: protocol,
-    ...modelLimits,
-    ...shared,
-    vision,
-    thinkingStyle: anthropic ? "anthropic" : "openai",
-    tokenField:
-      protocol === "responses" ? "max_completion_tokens" : "max_tokens",
-    reasoningEfforts:
-      reasoningEfforts ||
-      (anthropic
-        ? ["none", "low", "medium", "high"]
-        : ["none", "low", "medium", "high"]),
-    reasoningEffort,
-    temperature,
-    topP,
-    timeoutMs: 120000,
-  };
-}
-
-const OPENCODE_GO_MODELS = [
-  ["grok-4.7", "Grok 4.7", "responses"],
-  ["gpt-5.6-luna", "GPT-5.6 Luna", "responses", true],
-  ["deepseek-v4-pro", "DeepSeek V4 Pro", "chat"],
-  ["glm-5.3-flash", "GLM-5.3 Flash", "chat", true],
-  ["kimi-k3", "Kimi K3", "chat"],
-  ["minimax-m3", "MiniMax M3", "anthropic"],
-  ["qwen3.8-max", "Qwen3.8 Max", "anthropic", true],
-].map(([model, label, protocol, vision]) =>
-  openCodeModel({
-    channel: "go",
-    model,
-    label,
-    protocol,
-    vision: !!vision,
-    summary: `OpenCode Go · ${model} · ${protocol === "responses" ? "Responses" : protocol === "anthropic" ? "Messages" : "Chat Completions"} 接口。默认上下文和输出预算为保守起始值，可按账户实际模型上限调整。`,
-  }),
-);
-
-// Volcengine's Coding Plan catalog is intentionally separate from its normal
-// inference API: only the /api/coding/v3 endpoint consumes Coding Plan quota.
-function codingPlanModel({
-  model,
-  label,
-  context,
-  output,
-  vision = false,
-  summary = "",
-}) {
-  // Kimi K2.8 Preview advertises a 1M output ceiling. Reserve a practical
-  // 64K response budget here so the nearly 1M input window remains usable by
-  // LuckyTri; model-stage output limits are smaller for ordinary conversations.
-  const maxOutputTokens = Math.min(
-    output,
-    model === "kimi-k2.8-preview" ? 65536 : output,
-  );
-  const contextWindow = context;
-  return {
-    id: `volcengine-coding-plan-${model}`,
-    vendor: "火山方舟 Coding Plan",
-    label,
-    summary:
-      summary ||
-      `Coding Plan 专属 Responses 接口 · 上下文 ${context.toLocaleString("en-US")} · 官方最大输出 ${output.toLocaleString("en-US")}${vision ? " · 支持图片理解" : " · 文本模型"}。默认中等推理；Kimi K2.8 Preview 在 LuckyTri 中预留 64K 输出预算以保留长上下文。`,
-    provider: "volcengine-coding-plan",
-    baseUrl: VOLCENGINE_CODING_PLAN,
-    model,
-    apiProtocol: "responses",
-    ...limits(contextWindow, maxOutputTokens),
-    ...shared,
-    vision,
-    thinkingStyle: "volcengine",
-    tokenField: "max_completion_tokens",
-    reasoningEfforts: ["none", "minimal", "low", "medium", "high"],
-    reasoningEffort: "medium",
-    // Reasoning models may reject sampling parameters. The Responses API uses
-    // reasoning.effort; the "none" choice maps to minimal reasoning.
-    omitSampling: true,
-    temperature: 1,
-    topP: 1,
-    timeoutMs: 120000,
-  };
-}
-
-const VOLCENGINE_CODING_PLAN_MODELS = [
-  codingPlanModel({
-    model: "ark-code-latest",
-    label: "方舟自动路由（ark-code-latest）",
-    context: 256000,
-    output: 32768,
-    vision: true,
-    summary:
-      "控制台管理的动态模型入口；实际使用哪一个模型由 Coding Plan 控制台选择，切换通常数分钟后生效。支持图片理解。",
-  }),
-  codingPlanModel({
-    model: "doubao-seed-evolving",
-    label: "Doubao Seed Evolving",
-    context: 1024000,
-    output: 262144,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "doubao-seed-2.1-pro",
-    label: "Doubao Seed 2.1 Pro",
-    context: 1024000,
-    output: 262144,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "doubao-seed-2.1-lite",
-    label: "Doubao Seed 2.1 Lite",
-    context: 1024000,
-    output: 256000,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "doubao-seed-2.0-mini",
-    label: "Doubao Seed 2.0 Mini",
-    context: 256000,
-    output: 128000,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "minimax-m3",
-    label: "MiniMax M3",
-    context: 1024000,
-    output: 131072,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "glm-5.3",
-    label: "GLM 5.3",
-    context: 1024000,
-    output: 131072,
-  }),
-  codingPlanModel({
-    model: "glm-latest",
-    label: "GLM 最新别名（glm-latest）",
-    context: 1024000,
-    output: 131072,
-  }),
-  codingPlanModel({
-    model: "glm-5.3-flash",
-    label: "GLM 5.3 Flash",
-    context: 1024000,
-    output: 131072,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "deepseek-v4.1-flash",
-    label: "DeepSeek V4.1 Flash",
-    context: 1024000,
-    output: 393216,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "deepseek-v4-flash",
-    label: "DeepSeek V4 Flash",
-    context: 1024000,
-    output: 393216,
-  }),
-  codingPlanModel({
-    model: "deepseek-v4-pro",
-    label: "DeepSeek V4 Pro",
-    context: 1024000,
-    output: 393216,
-  }),
-  codingPlanModel({
-    model: "kimi-k2.7-code",
-    label: "Kimi K2.7 Code",
-    context: 256000,
-    output: 32768,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "kimi-k2.8-preview",
-    label: "Kimi K2.8 Preview",
-    context: 1024000,
-    output: 1024000,
-    vision: true,
-  }),
-  codingPlanModel({
-    model: "kimi-k3",
-    label: "Kimi K3",
-    context: 1024000,
-    output: 131072,
-    vision: true,
-  }),
-];
-
-export const EFFORT_LABELS = {
-  none: "关闭",
-  minimal: "极低",
-  low: "低",
-  medium: "中",
-  high: "高",
-  xhigh: "极高",
-  max: "最深",
-};
-
-export const MODEL_CATALOG = [
-  ...VOLCENGINE_CODING_PLAN_MODELS,
-  {
-    id: "deepseek-flash",
-    vendor: "DeepSeek",
-    label: "DeepSeek V4.1 Flash",
-    summary:
-      "当前主力。思考默认开启（high），可关闭；支持看图。1M 内同一价格，没有分档。",
-    provider: "deepseek",
-    baseUrl: "https://api.deepseek.com",
-    model: "deepseek-flash",
-    ...limits(1000000, 393216),
-    ...shared,
-    vision: true,
-    thinkingStyle: "deepseek",
-    tokenField: "max_tokens",
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    temperature: 1,
-    topP: 1,
-  },
-  {
-    id: "deepseek-v4-pro",
-    vendor: "DeepSeek",
-    label: "DeepSeek V4 Pro",
-    summary:
-      "V4-Pro-0813。思考默认开启（high），可关闭；仅文本。1M 内同一价格，没有分档。",
-    provider: "deepseek",
-    baseUrl: "https://api.deepseek.com",
-    model: "deepseek-v4-pro",
-    ...limits(1000000, 393216),
-    ...shared,
-    vision: false,
-    thinkingStyle: "deepseek",
-    tokenField: "max_tokens",
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    temperature: 1,
-    topP: 1,
-  },
-  {
-    id: "mimo-v2.6-pro",
-    vendor: "小米 MiMo",
-    label: "MiMo V2.6 Pro",
-    summary:
-      "旗舰。思考默认开启，可关闭；支持图片、视频、音频输入。1M 内同一价格，没有分档。",
-    provider: "mimo",
-    baseUrl: "https://api.xiaomimimo.com/v1",
-    model: "mimo-v2.6-pro",
-    ...limits(1048576, 131072),
-    ...shared,
-    vision: true,
-    thinkingStyle: "mimo",
-    tokenField: "max_completion_tokens",
-    reasoningEfforts: ["none", "high"],
-    reasoningEffort: "high",
-    temperature: 1,
-    topP: 0.95,
-  },
-  {
-    id: "mimo-v2.6-flash",
-    vendor: "小米 MiMo",
-    label: "MiMo V2.6 Flash",
-    summary:
-      "高频调用。思考默认开启，可关闭；支持图片、视频、音频输入。1M 内同一价格，没有分档。",
-    provider: "mimo",
-    baseUrl: "https://api.xiaomimimo.com/v1",
-    model: "mimo-v2.6-flash",
-    ...limits(1048576, 131072),
-    ...shared,
-    vision: true,
-    thinkingStyle: "mimo",
-    tokenField: "max_completion_tokens",
-    reasoningEfforts: ["none", "high"],
-    reasoningEffort: "high",
-    temperature: 1,
-    topP: 0.95,
-  },
-  {
-    id: "glm-5.3",
-    vendor: "智谱 GLM",
-    label: "GLM-5.3",
-    summary: "旗舰。思考不能关闭，默认 max；仅文本。1M 内同一价格，没有分档。",
-    provider: "glm",
-    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
-    model: "glm-5.3",
-    ...limits(1000000, 131072),
-    ...shared,
-    vision: false,
-    thinkingStyle: "glm",
-    tokenField: "max_tokens",
-    reasoningEfforts: ["low", "high", "max"],
-    reasoningEffort: "max",
-    temperature: 1,
-    topP: 0.95,
-  },
-  {
-    id: "glm-5.3-flash",
-    vendor: "智谱 GLM",
-    label: "GLM-5.3 Flash",
-    summary:
-      "原生多模态。思考不能关闭，默认 max；支持图片、视频。1M 内同一价格，没有分档。",
-    provider: "glm",
-    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
-    model: "glm-5.3-flash",
-    ...limits(1000000, 131072),
-    ...shared,
-    vision: true,
-    thinkingStyle: "glm",
-    tokenField: "max_tokens",
-    reasoningEfforts: ["low", "high", "max"],
-    reasoningEffort: "max",
-    temperature: 1,
-    topP: 0.95,
-  },
-  {
-    id: "kimi-k3",
-    vendor: "Kimi",
-    label: "Kimi K3",
-    summary:
-      "旗舰。思考不能关闭，默认 max；支持看图。1M 内同一价格；输出和输入共用窗口，官方默认输出 128K。",
-    provider: "kimi",
-    baseUrl: "https://api.moonshot.cn/v1",
-    model: "kimi-k3",
-    ...limits(1048576, 131072),
-    ...shared,
-    vision: true,
-    thinkingStyle: "kimi-effort",
-    tokenField: "max_completion_tokens",
-    omitSampling: true,
-    reasoningEfforts: ["low", "high", "max"],
-    reasoningEffort: "max",
-    temperature: 1,
-    topP: 0.95,
-  },
-  {
-    id: "kimi-k2.6",
-    vendor: "Kimi",
-    label: "Kimi K2.6",
-    summary:
-      "通用模型。思考默认开启，可关闭；支持看图。输出和输入共用 256K 窗口，官方默认输出 32K。",
-    provider: "kimi",
-    baseUrl: "https://api.moonshot.cn/v1",
-    model: "kimi-k2.6",
-    ...limits(262144, 32768),
-    ...shared,
-    vision: true,
-    thinkingStyle: "kimi-toggle",
-    tokenField: "max_completion_tokens",
-    omitSampling: true,
-    reasoningEfforts: ["none", "high"],
-    reasoningEffort: "high",
-    temperature: 1,
-    topP: 0.95,
-  },
-  gpt({
-    id: "gpt-6-astra",
-    vendor: "OpenAI",
-    label: "GPT-6 Astra",
-    summary:
-      "最强旗舰。思考不能关闭，官方未写默认档，预填 medium；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "openai",
-    baseUrl: OPENAI,
-    model: "gpt-6-astra",
-    reasoningEfforts: ASTRA_EFFORTS,
-  }),
-  gpt({
-    id: "gpt-6-sol",
-    vendor: "OpenAI",
-    label: "GPT-6 Sol",
-    summary:
-      "复杂编码与智能体。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "openai",
-    baseUrl: OPENAI,
-    model: "gpt-6-sol",
-  }),
-  gpt({
-    id: "gpt-6-luna",
-    vendor: "OpenAI",
-    label: "GPT-6 Luna",
-    summary:
-      "高频轻量。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "openai",
-    baseUrl: OPENAI,
-    model: "gpt-6-luna",
-  }),
-  openRouterGpt({
-    id: "openrouter-gpt-6-astra",
-    vendor: "OpenRouter",
-    label: "GPT-6 Astra",
-    summary:
-      "OpenRouter 路由 · openai/gpt-6-astra。1.05M 上下文、128K 最大输出；支持图片、结构化输出和工具调用。",
-    model: "openai/gpt-6-astra",
-    reasoningEfforts: ASTRA_EFFORTS,
-  }),
-  openRouterGpt({
-    id: "openrouter-gpt-6-sol",
-    vendor: "OpenRouter",
-    label: "GPT-6 Sol",
-    summary:
-      "OpenRouter 路由 · openai/gpt-6-sol。1.05M 上下文、128K 最大输出；支持图片、结构化输出和工具调用。",
-    model: "openai/gpt-6-sol",
-  }),
-  openRouterGpt({
-    id: "openrouter-gpt-6-luna",
-    vendor: "OpenRouter",
-    label: "GPT-6 Luna",
-    summary:
-      "OpenRouter 路由 · openai/gpt-6-luna。1.05M 上下文、128K 最大输出；支持图片、结构化输出和工具调用。",
-    model: "openai/gpt-6-luna",
-  }),
-  {
-    id: "openrouter-glm-5.3-flash",
-    vendor: "OpenRouter",
-    label: "GLM-5.3 Flash",
-    summary:
-      "OpenRouter 路由 · z-ai/glm-5.3-flash。1,310,720 上下文、131,072 最大输出；支持图片、结构化输出和工具调用。",
-    provider: "openrouter",
-    baseUrl: OPENROUTER,
-    model: "z-ai/glm-5.3-flash",
-    ...limits(1310720, 131072),
-    ...shared,
-    vision: true,
-    thinkingStyle: "openai",
-    tokenField: "max_tokens",
-    reasoningEfforts: ["low", "high", "max"],
-    reasoningEffort: "max",
-    temperature: 1,
-    topP: 0.95,
-  },
-  gpt({
-    id: "gpt-5.6-sol",
-    vendor: "OpenAI",
-    label: "GPT-5.6 Sol",
-    summary:
-      "上一代旗舰，别名 gpt-5.6。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "openai",
-    baseUrl: OPENAI,
-    model: "gpt-5.6-sol",
-  }),
-  gpt({
-    id: "gpt-5.6-terra",
-    vendor: "OpenAI",
-    label: "GPT-5.6 Terra",
-    summary:
-      "上一代均衡款。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "openai",
-    baseUrl: OPENAI,
-    model: "gpt-5.6-terra",
-  }),
-  gpt({
-    id: "gpt-5.6-luna",
-    vendor: "OpenAI",
-    label: "GPT-5.6 Luna",
-    summary:
-      "上一代高性价比。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "openai",
-    baseUrl: OPENAI,
-    model: "gpt-5.6-luna",
-  }),
-  {
-    id: "qwen3.8-max",
-    vendor: "通义千问",
-    label: "Qwen3.8 Max",
-    summary:
-      "旗舰，快照 qwen3.8-max-0902。思考默认开启（xhigh），可关闭；支持图片、视频。1M 内同一价格，没有分档。",
-    provider: "qwen",
-    baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
-    model: "qwen3.8-max",
-    ...limits(1000000, 131072),
-    ...shared,
-    vision: true,
-    thinkingStyle: "qwen-effort",
-    tokenField: "max_completion_tokens",
-    omitSampling: true,
-    reasoningEfforts: ["none", "low", "medium", "xhigh"],
-    reasoningEffort: "xhigh",
-    temperature: 0.6,
-    topP: 0.95,
-  },
-  {
-    id: "qwen3.8-flash",
-    vendor: "通义千问",
-    label: "Qwen3.8 Flash",
-    summary:
-      "轻量。思考默认开启（xhigh），可关闭；支持图片、视频。1M 内同一价格，没有分档。",
-    provider: "qwen",
-    baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
-    model: "qwen3.8-flash",
-    ...limits(1000000, 131072),
-    ...shared,
-    vision: true,
-    thinkingStyle: "qwen-effort",
-    tokenField: "max_completion_tokens",
-    omitSampling: true,
-    reasoningEfforts: ["none", "low", "medium", "xhigh"],
-    reasoningEffort: "xhigh",
-    temperature: 0.6,
-    topP: 0.95,
-  },
-  gpt({
-    id: "bedrock-gpt-6-astra",
-    vendor: "亚马逊 Bedrock",
-    label: "GPT-6 Astra",
-    summary:
-      "美国跨区推理。思考不能关闭，预填 medium；支持看图。runtime 不支持结构化输出，已关闭 JSON 输出。输入超过 272K 整单按长上下文计价。",
-    provider: "bedrock",
-    baseUrl: BEDROCK,
-    model: "us.openai.gpt-6-astra",
-    json: false,
-    reasoningEfforts: ASTRA_EFFORTS,
-  }),
-  gpt({
-    id: "bedrock-gpt-6-sol",
-    vendor: "亚马逊 Bedrock",
-    label: "GPT-6 Sol",
-    summary:
-      "AWS 模型卡尚未发布，模型 ID 取自 OpenAI 的 Bedrock 指南，JSON 输出先关闭。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "bedrock",
-    baseUrl: BEDROCK,
-    model: "us.openai.gpt-6-sol",
-    json: false,
-  }),
-  gpt({
-    id: "bedrock-gpt-6-luna",
-    vendor: "亚马逊 Bedrock",
-    label: "GPT-6 Luna",
-    summary:
-      "AWS 模型卡尚未发布，模型 ID 取自 OpenAI 的 Bedrock 指南，JSON 输出先关闭。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
-    provider: "bedrock",
-    baseUrl: BEDROCK,
-    model: "us.openai.gpt-6-luna",
-    json: false,
-  }),
-  ...OPENCODE_GO_MODELS,
-  gpt({
-    id: "opencode-zen-gpt-6-astra",
-    vendor: "OpenCode Zen",
-    label: "GPT-6 Astra",
-    summary: "OpenCode Zen · GPT-6 系列旗舰 · Responses 接口。",
-    provider: "opencode-zen",
-    baseUrl: OPENCODE_ZEN,
-    model: "gpt-6-astra",
-    apiProtocol: "responses",
-    reasoningEfforts: ASTRA_EFFORTS,
-  }),
-  gpt({
-    id: "opencode-zen-gpt-6-sol",
-    vendor: "OpenCode Zen",
-    label: "GPT-6 Sol",
-    summary: "OpenCode Zen · GPT-6 均衡款 · Responses 接口。",
-    provider: "opencode-zen",
-    baseUrl: OPENCODE_ZEN,
-    model: "gpt-6-sol",
-    apiProtocol: "responses",
-  }),
-  gpt({
-    id: "opencode-zen-gpt-6-luna",
-    vendor: "OpenCode Zen",
-    label: "GPT-6 Luna",
-    summary: "OpenCode Zen · GPT-6 轻量款 · Responses 接口。",
-    provider: "opencode-zen",
-    baseUrl: OPENCODE_ZEN,
-    model: "gpt-6-luna",
-    apiProtocol: "responses",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "deepseek-v4.1-flash",
-    label: "DeepSeek V4.1 Flash",
-    limits: limits(1000000, 393216),
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    summary: "OpenCode Zen · 快速推理 · Chat Completions 接口。",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "deepseek-v4-pro",
-    label: "DeepSeek V4 Pro",
-    limits: limits(1000000, 393216),
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    summary: "OpenCode Zen · DeepSeek V4 旗舰 · Chat Completions 接口。",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "deepseek-v4-flash",
-    label: "DeepSeek V4 Flash",
-    limits: limits(1000000, 393216),
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    summary: "OpenCode Zen · 轻量高速 · Chat Completions 接口。",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "glm-5.3-flash",
-    label: "GLM-5.3 Flash",
-    vision: true,
-    limits: limits(1000000, 131072),
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    summary: "OpenCode Zen · 智谱轻量多模态 · Chat Completions 接口。",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "glm-5.3",
-    label: "GLM-5.3",
-    limits: limits(1000000, 131072),
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    summary: "OpenCode Zen · 智谱旗舰 · Chat Completions 接口。",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "kimi-k3",
-    label: "Kimi K3",
-    vision: true,
-    limits: limits(1048576, 131072),
-    reasoningEfforts: ["none", "low", "high", "max"],
-    reasoningEffort: "high",
-    summary: "OpenCode Zen · Kimi 旗舰 · Chat Completions 接口。",
-  }),
-  // Qwen3.8 Max is only served through Go; Zen offers Qwen3.8 Flash.
-  openCodeModel({
-    channel: "zen",
-    model: "qwen3.8-flash",
-    label: "Qwen3.8 Flash",
-    protocol: "anthropic",
-    vision: true,
-    limits: limits(1000000, 131072),
-    summary: "OpenCode Zen · 通义千问轻量多模态 · Anthropic Messages 接口。",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "minimax-m3",
-    label: "MiniMax M3",
-    limits: limits(128000, 8192),
-    reasoningEfforts: ["none", "low", "medium", "high"],
-    summary: "OpenCode Zen · MiniMax 主力模型 · Chat Completions 接口。",
-  }),
-  openCodeModel({
-    channel: "zen",
-    model: "space-bunny-free",
-    label: "Space Bunny Free",
-    limits: {
-      contextWindow: 1000000,
-      maxInputTokens: 128000,
-      maxOutputTokens: 8192,
-    },
-    vision: true,
-    reasoningEfforts: [
-      "none",
-      "minimal",
-      "low",
-      "medium",
-      "high",
-      "xhigh",
-      "max",
-    ],
-    reasoningEffort: "low",
-    summary:
-      "OpenCode Zen · 当前限时免费 · Chat Completions 接口。参考 OpenRouter Space Bunny Alpha 规格：1M 上下文、最高 524K 输出，支持图像/视频、推理、工具和 JSON；其匿名第三方提供方可能保留提示和回答。LuckyTri 默认输入 128K、输出 8K、低推理；当前图片链路可用，视频链路未支持。Zen 路由能力与数据政策可能不同。",
-  }),
-  {
-    id: "custom",
-    vendor: "自定义",
-    label: "自定义兼容接口",
-    summary: "OpenAI 兼容接口。上下文先按 128K 填，思考档可按供应商再改。",
-    provider: "custom",
-    baseUrl: "",
-    model: "",
-    ...limits(128000, 8192),
-    embedding: false,
-    embeddingModel: "",
-    timeoutMs: 90000,
-    system: true,
-    json: true,
-    tools: false,
-    vision: false,
-    thinkingStyle: "openai",
-    tokenField: "max_tokens",
-    reasoningEfforts: [
-      "none",
-      "minimal",
-      "low",
-      "medium",
-      "high",
-      "xhigh",
-      "max",
-    ],
-    reasoningEffort: "none",
-    temperature: 0.85,
-    topP: 1,
-  },
-  {
-    id: "openrouter",
-    vendor: "OpenRouter",
-    label: "OpenRouter · 自选模型",
-    summary:
-      "已填好 OpenRouter API 地址。请自行填写模型 ID、API Key，并按所选模型设置上下文、输出和能力参数。",
-    provider: "openrouter",
-    baseUrl: OPENROUTER,
-    model: "",
-    ...limits(128000, 8192),
-    embedding: false,
-    embeddingModel: "",
-    timeoutMs: 90000,
-    system: true,
-    json: true,
-    tools: false,
-    vision: false,
-    thinkingStyle: "openai",
-    tokenField: "max_tokens",
-    reasoningEfforts: [
-      "none",
-      "minimal",
-      "low",
-      "medium",
-      "high",
-      "xhigh",
-      "max",
-    ],
-    reasoningEffort: "none",
-    temperature: 0.85,
-    topP: 1,
-  },
-];
-
-export const MODEL_PRESETS = {
-  custom: { label: "自定义兼容接口", baseUrl: "", model: "" },
-  deepseek: {
-    label: "DeepSeek",
-    baseUrl: "https://api.deepseek.com",
-    model: "deepseek-flash",
-  },
-  openai: {
-    label: "OpenAI",
-    baseUrl: OPENAI,
-    model: "gpt-6-astra",
-  },
-  bedrock: {
-    label: "亚马逊 Bedrock",
-    baseUrl: BEDROCK,
-    model: "us.openai.gpt-6-astra",
-  },
-  qwen: {
-    label: "通义千问",
-    baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
-    model: "qwen3.8-max",
-  },
-  moonshot: {
-    label: "Kimi",
-    baseUrl: "https://api.moonshot.cn/v1",
-    model: "kimi-k3",
-  },
-  kimi: {
-    label: "Kimi",
-    baseUrl: "https://api.moonshot.cn/v1",
-    model: "kimi-k3",
-  },
-  zhipu: {
-    label: "智谱 GLM",
-    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
-    model: "glm-5.3",
-  },
-  glm: {
-    label: "智谱 GLM",
-    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
-    model: "glm-5.3",
-  },
-  mimo: {
-    label: "小米 MiMo",
-    baseUrl: "https://api.xiaomimimo.com/v1",
-    model: "mimo-v2.6-pro",
-  },
-  siliconflow: {
-    label: "SiliconFlow",
-    baseUrl: "https://api.siliconflow.cn/v1",
-    model: "",
-  },
-  openrouter: {
-    label: "OpenRouter",
-    baseUrl: OPENROUTER,
-    model: "",
-  },
-  "volcengine-coding-plan": {
-    label: "火山方舟 Coding Plan",
-    baseUrl: VOLCENGINE_CODING_PLAN,
-    model: "ark-code-latest",
-  },
-};
-
-export const REASONING_EFFORTS = [
-  { id: "none", label: "关闭" },
-  { id: "minimal", label: "极低" },
-  { id: "low", label: "低" },
-  { id: "medium", label: "中" },
-  { id: "high", label: "高" },
-  { id: "xhigh", label: "极高" },
-  { id: "max", label: "最深" },
-];
+// Checked against vendor API docs on 2026-09-23.
+// maxOutputTokens is the published output ceiling and maxInputTokens is what
+// the window leaves after it, so every preset passes validateModel as is.
+
+function limits(contextWindow, maxOutputTokens) {
+  return {
+    contextWindow,
+    maxOutputTokens,
+    maxInputTokens: contextWindow - maxOutputTokens,
+  };
+}
+
+// OpenAI and Bedrock bill the whole GPT request at the long-context rate once
+// input passes 272K tokens. The standard tier stops input there and keeps the
+// full 128K output.
+function gptWindows() {
+  const standard = {
+    contextWindow: 272000 + 128000,
+    maxInputTokens: 272000,
+    maxOutputTokens: 128000,
+  };
+  return {
+    ...standard,
+    contextWindows: [
+      { label: "标准（输入 ≤272K）", ...standard },
+      { label: "百万（1.05M）", ...limits(1050000, 128000) },
+    ],
+  };
+}
+
+const shared = {
+  embedding: false,
+  embeddingModel: "",
+  timeoutMs: 120000,
+  system: true,
+  json: true,
+  tools: true,
+};
+
+const OPENAI = "https://api.openai.com/v1";
+const BEDROCK = "https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1";
+const OPENROUTER = "https://openrouter.ai/api/v1";
+const OPENCODE_GO = "https://opencode.ai/zen/go/v1";
+const OPENCODE_ZEN = "https://opencode.ai/zen/v1";
+const VOLCENGINE_CODING_PLAN =
+  "https://ark.cn-beijing.volces.com/api/coding/v3";
+const GPT_EFFORTS = ["none", "low", "medium", "high", "xhigh", "max"];
+const ASTRA_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
+
+function gpt(fields) {
+  return {
+    ...gptWindows(),
+    ...shared,
+    vision: true,
+    thinkingStyle: "openai",
+    tokenField: "max_completion_tokens",
+    reasoningEfforts: GPT_EFFORTS,
+    reasoningEffort: "medium",
+    temperature: 1,
+    topP: 1,
+    ...fields,
+  };
+}
+
+function openRouterGpt(fields) {
+  const profile = gpt({
+    ...fields,
+    provider: "openrouter",
+    baseUrl: OPENROUTER,
+    contextWindow: 1050000,
+    maxInputTokens: 922000,
+    maxOutputTokens: 128000,
+  });
+  delete profile.contextWindows;
+  return profile;
+}
+
+function openCodeModel({
+  channel,
+  model,
+  label,
+  protocol = "chat",
+  summary,
+  vision = false,
+  reasoningEfforts,
+  reasoningEffort = "none",
+  limits: modelLimits = limits(128000, 8192),
+  temperature = 0.85,
+  topP = 1,
+}) {
+  const isGo = channel === "go";
+  const anthropic = protocol === "anthropic";
+  return {
+    id: `opencode-${channel}-${model}`,
+    vendor: isGo ? "OpenCode Go" : "OpenCode Zen",
+    label,
+    summary,
+    provider: isGo ? "opencode-go" : "opencode-zen",
+    baseUrl: isGo ? OPENCODE_GO : OPENCODE_ZEN,
+    model,
+    apiProtocol: protocol,
+    ...modelLimits,
+    ...shared,
+    vision,
+    thinkingStyle: anthropic ? "anthropic" : "openai",
+    tokenField:
+      protocol === "responses" ? "max_completion_tokens" : "max_tokens",
+    reasoningEfforts:
+      reasoningEfforts ||
+      (anthropic
+        ? ["none", "low", "medium", "high"]
+        : ["none", "low", "medium", "high"]),
+    reasoningEffort,
+    temperature,
+    topP,
+    timeoutMs: 120000,
+  };
+}
+
+const OPENCODE_GO_MODELS = [
+  ["grok-4.7", "Grok 4.7", "responses"],
+  ["gpt-5.6-luna", "GPT-5.6 Luna", "responses", true],
+  ["deepseek-v4-pro", "DeepSeek V4 Pro", "chat"],
+  ["glm-5.3-flash", "GLM-5.3 Flash", "chat", true],
+  ["kimi-k3", "Kimi K3", "chat"],
+  ["minimax-m3", "MiniMax M3", "anthropic"],
+  ["qwen3.8-max", "Qwen3.8 Max", "anthropic", true],
+].map(([model, label, protocol, vision]) =>
+  openCodeModel({
+    channel: "go",
+    model,
+    label,
+    protocol,
+    vision: !!vision,
+    summary: `OpenCode Go · ${model} · ${protocol === "responses" ? "Responses" : protocol === "anthropic" ? "Messages" : "Chat Completions"} 接口。默认上下文和输出预算为保守起始值，可按账户实际模型上限调整。`,
+  }),
+);
+
+// Volcengine's Coding Plan catalog is intentionally separate from its normal
+// inference API: only the /api/coding/v3 endpoint consumes Coding Plan quota.
+function codingPlanModel({
+  model,
+  label,
+  context,
+  output,
+  vision = false,
+  summary = "",
+}) {
+  // Kimi K2.8 Preview advertises a 1M output ceiling. Reserve a practical
+  // 64K response budget here so the nearly 1M input window remains usable by
+  // LuckyTri; model-stage output limits are smaller for ordinary conversations.
+  const maxOutputTokens = Math.min(
+    output,
+    model === "kimi-k2.8-preview" ? 65536 : output,
+  );
+  const contextWindow = context;
+  return {
+    id: `volcengine-coding-plan-${model}`,
+    vendor: "火山方舟 Coding Plan",
+    label,
+    summary:
+      summary ||
+      `Coding Plan 专属 Responses 接口 · 上下文 ${context.toLocaleString("en-US")} · 官方最大输出 ${output.toLocaleString("en-US")}${vision ? " · 支持图片理解" : " · 文本模型"}。默认中等推理；Kimi K2.8 Preview 在 LuckyTri 中预留 64K 输出预算以保留长上下文。`,
+    provider: "volcengine-coding-plan",
+    baseUrl: VOLCENGINE_CODING_PLAN,
+    model,
+    apiProtocol: "responses",
+    ...limits(contextWindow, maxOutputTokens),
+    ...shared,
+    vision,
+    thinkingStyle: "volcengine",
+    tokenField: "max_completion_tokens",
+    reasoningEfforts: ["none", "minimal", "low", "medium", "high"],
+    reasoningEffort: "medium",
+    // Reasoning models may reject sampling parameters. The Responses API uses
+    // reasoning.effort; the "none" choice maps to minimal reasoning.
+    omitSampling: true,
+    temperature: 1,
+    topP: 1,
+    timeoutMs: 120000,
+  };
+}
+
+const VOLCENGINE_CODING_PLAN_MODELS = [
+  codingPlanModel({
+    model: "ark-code-latest",
+    label: "方舟自动路由（ark-code-latest）",
+    context: 256000,
+    output: 32768,
+    vision: true,
+    summary:
+      "控制台管理的动态模型入口；实际使用哪一个模型由 Coding Plan 控制台选择，切换通常数分钟后生效。支持图片理解。",
+  }),
+  codingPlanModel({
+    model: "doubao-seed-evolving",
+    label: "Doubao Seed Evolving",
+    context: 1024000,
+    output: 262144,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "doubao-seed-2.1-pro",
+    label: "Doubao Seed 2.1 Pro",
+    context: 1024000,
+    output: 262144,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "doubao-seed-2.1-lite",
+    label: "Doubao Seed 2.1 Lite",
+    context: 1024000,
+    output: 256000,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "doubao-seed-2.0-mini",
+    label: "Doubao Seed 2.0 Mini",
+    context: 256000,
+    output: 128000,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "minimax-m3",
+    label: "MiniMax M3",
+    context: 1024000,
+    output: 131072,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "glm-5.3",
+    label: "GLM 5.3",
+    context: 1024000,
+    output: 131072,
+  }),
+  codingPlanModel({
+    model: "glm-latest",
+    label: "GLM 最新别名（glm-latest）",
+    context: 1024000,
+    output: 131072,
+  }),
+  codingPlanModel({
+    model: "glm-5.3-flash",
+    label: "GLM 5.3 Flash",
+    context: 1024000,
+    output: 131072,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "deepseek-v4.1-flash",
+    label: "DeepSeek V4.1 Flash",
+    context: 1024000,
+    output: 393216,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "deepseek-v4-flash",
+    label: "DeepSeek V4 Flash",
+    context: 1024000,
+    output: 393216,
+  }),
+  codingPlanModel({
+    model: "deepseek-v4-pro",
+    label: "DeepSeek V4 Pro",
+    context: 1024000,
+    output: 393216,
+  }),
+  codingPlanModel({
+    model: "kimi-k2.7-code",
+    label: "Kimi K2.7 Code",
+    context: 256000,
+    output: 32768,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "kimi-k2.8-preview",
+    label: "Kimi K2.8 Preview",
+    context: 1024000,
+    output: 1024000,
+    vision: true,
+  }),
+  codingPlanModel({
+    model: "kimi-k3",
+    label: "Kimi K3",
+    context: 1024000,
+    output: 131072,
+    vision: true,
+  }),
+];
+
+export const EFFORT_LABELS = {
+  none: "关闭",
+  minimal: "极低",
+  low: "低",
+  medium: "中",
+  high: "高",
+  xhigh: "极高",
+  max: "最深",
+};
+
+export const MODEL_CATALOG = [
+  ...VOLCENGINE_CODING_PLAN_MODELS,
+  {
+    id: "deepseek-flash",
+    vendor: "DeepSeek",
+    label: "DeepSeek V4.1 Flash",
+    summary:
+      "当前主力。思考默认开启（high），可关闭；支持看图。1M 内同一价格，没有分档。",
+    provider: "deepseek",
+    baseUrl: "https://api.deepseek.com",
+    model: "deepseek-flash",
+    ...limits(1000000, 393216),
+    ...shared,
+    vision: true,
+    thinkingStyle: "deepseek",
+    tokenField: "max_tokens",
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    temperature: 1,
+    topP: 1,
+  },
+  {
+    id: "deepseek-v4-pro",
+    vendor: "DeepSeek",
+    label: "DeepSeek V4 Pro",
+    summary:
+      "V4-Pro-0813。思考默认开启（high），可关闭；仅文本。1M 内同一价格，没有分档。",
+    provider: "deepseek",
+    baseUrl: "https://api.deepseek.com",
+    model: "deepseek-v4-pro",
+    ...limits(1000000, 393216),
+    ...shared,
+    vision: false,
+    thinkingStyle: "deepseek",
+    tokenField: "max_tokens",
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    temperature: 1,
+    topP: 1,
+  },
+  {
+    id: "mimo-v2.6-pro",
+    vendor: "小米 MiMo",
+    label: "MiMo V2.6 Pro",
+    summary:
+      "旗舰。思考默认开启，可关闭；支持图片、视频、音频输入。1M 内同一价格，没有分档。",
+    provider: "mimo",
+    baseUrl: "https://api.xiaomimimo.com/v1",
+    model: "mimo-v2.6-pro",
+    ...limits(1048576, 131072),
+    ...shared,
+    vision: true,
+    thinkingStyle: "mimo",
+    tokenField: "max_completion_tokens",
+    reasoningEfforts: ["none", "high"],
+    reasoningEffort: "high",
+    temperature: 1,
+    topP: 0.95,
+  },
+  {
+    id: "mimo-v2.6-flash",
+    vendor: "小米 MiMo",
+    label: "MiMo V2.6 Flash",
+    summary:
+      "高频调用。思考默认开启，可关闭；支持图片、视频、音频输入。1M 内同一价格，没有分档。",
+    provider: "mimo",
+    baseUrl: "https://api.xiaomimimo.com/v1",
+    model: "mimo-v2.6-flash",
+    ...limits(1048576, 131072),
+    ...shared,
+    vision: true,
+    thinkingStyle: "mimo",
+    tokenField: "max_completion_tokens",
+    reasoningEfforts: ["none", "high"],
+    reasoningEffort: "high",
+    temperature: 1,
+    topP: 0.95,
+  },
+  {
+    id: "glm-5.3",
+    vendor: "智谱 GLM",
+    label: "GLM-5.3",
+    summary: "旗舰。思考不能关闭，默认 max；仅文本。1M 内同一价格，没有分档。",
+    provider: "glm",
+    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
+    model: "glm-5.3",
+    ...limits(1000000, 131072),
+    ...shared,
+    vision: false,
+    thinkingStyle: "glm",
+    tokenField: "max_tokens",
+    reasoningEfforts: ["low", "high", "max"],
+    reasoningEffort: "max",
+    temperature: 1,
+    topP: 0.95,
+  },
+  {
+    id: "glm-5.3-flash",
+    vendor: "智谱 GLM",
+    label: "GLM-5.3 Flash",
+    summary:
+      "原生多模态。思考不能关闭，默认 max；支持图片、视频。1M 内同一价格，没有分档。",
+    provider: "glm",
+    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
+    model: "glm-5.3-flash",
+    ...limits(1000000, 131072),
+    ...shared,
+    vision: true,
+    thinkingStyle: "glm",
+    tokenField: "max_tokens",
+    reasoningEfforts: ["low", "high", "max"],
+    reasoningEffort: "max",
+    temperature: 1,
+    topP: 0.95,
+  },
+  {
+    id: "kimi-k3",
+    vendor: "Kimi",
+    label: "Kimi K3",
+    summary:
+      "旗舰。思考不能关闭，默认 max；支持看图。1M 内同一价格；输出和输入共用窗口，官方默认输出 128K。",
+    provider: "kimi",
+    baseUrl: "https://api.moonshot.cn/v1",
+    model: "kimi-k3",
+    ...limits(1048576, 131072),
+    ...shared,
+    vision: true,
+    thinkingStyle: "kimi-effort",
+    tokenField: "max_completion_tokens",
+    omitSampling: true,
+    reasoningEfforts: ["low", "high", "max"],
+    reasoningEffort: "max",
+    temperature: 1,
+    topP: 0.95,
+  },
+  {
+    id: "kimi-k2.6",
+    vendor: "Kimi",
+    label: "Kimi K2.6",
+    summary:
+      "通用模型。思考默认开启，可关闭；支持看图。输出和输入共用 256K 窗口，官方默认输出 32K。",
+    provider: "kimi",
+    baseUrl: "https://api.moonshot.cn/v1",
+    model: "kimi-k2.6",
+    ...limits(262144, 32768),
+    ...shared,
+    vision: true,
+    thinkingStyle: "kimi-toggle",
+    tokenField: "max_completion_tokens",
+    omitSampling: true,
+    reasoningEfforts: ["none", "high"],
+    reasoningEffort: "high",
+    temperature: 1,
+    topP: 0.95,
+  },
+  gpt({
+    id: "gpt-6-astra",
+    vendor: "OpenAI",
+    label: "GPT-6 Astra",
+    summary:
+      "最强旗舰。思考不能关闭，官方未写默认档，预填 medium；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "openai",
+    baseUrl: OPENAI,
+    model: "gpt-6-astra",
+    reasoningEfforts: ASTRA_EFFORTS,
+  }),
+  gpt({
+    id: "gpt-6-sol",
+    vendor: "OpenAI",
+    label: "GPT-6 Sol",
+    summary:
+      "复杂编码与智能体。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "openai",
+    baseUrl: OPENAI,
+    model: "gpt-6-sol",
+  }),
+  gpt({
+    id: "gpt-6-luna",
+    vendor: "OpenAI",
+    label: "GPT-6 Luna",
+    summary:
+      "高频轻量。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "openai",
+    baseUrl: OPENAI,
+    model: "gpt-6-luna",
+  }),
+  openRouterGpt({
+    id: "openrouter-gpt-6-astra",
+    vendor: "OpenRouter",
+    label: "GPT-6 Astra",
+    summary:
+      "OpenRouter 路由 · openai/gpt-6-astra。1.05M 上下文、128K 最大输出；支持图片、结构化输出和工具调用。",
+    model: "openai/gpt-6-astra",
+    reasoningEfforts: ASTRA_EFFORTS,
+  }),
+  openRouterGpt({
+    id: "openrouter-gpt-6-sol",
+    vendor: "OpenRouter",
+    label: "GPT-6 Sol",
+    summary:
+      "OpenRouter 路由 · openai/gpt-6-sol。1.05M 上下文、128K 最大输出；支持图片、结构化输出和工具调用。",
+    model: "openai/gpt-6-sol",
+  }),
+  openRouterGpt({
+    id: "openrouter-gpt-6-luna",
+    vendor: "OpenRouter",
+    label: "GPT-6 Luna",
+    summary:
+      "OpenRouter 路由 · openai/gpt-6-luna。1.05M 上下文、128K 最大输出；支持图片、结构化输出和工具调用。",
+    model: "openai/gpt-6-luna",
+  }),
+  {
+    id: "openrouter-glm-5.3-flash",
+    vendor: "OpenRouter",
+    label: "GLM-5.3 Flash",
+    summary:
+      "OpenRouter 路由 · z-ai/glm-5.3-flash。1,310,720 上下文、131,072 最大输出；支持图片、结构化输出和工具调用。",
+    provider: "openrouter",
+    baseUrl: OPENROUTER,
+    model: "z-ai/glm-5.3-flash",
+    ...limits(1310720, 131072),
+    ...shared,
+    vision: true,
+    thinkingStyle: "openai",
+    tokenField: "max_tokens",
+    reasoningEfforts: ["low", "high", "max"],
+    reasoningEffort: "max",
+    temperature: 1,
+    topP: 0.95,
+  },
+  gpt({
+    id: "gpt-5.6-sol",
+    vendor: "OpenAI",
+    label: "GPT-5.6 Sol",
+    summary:
+      "上一代旗舰，别名 gpt-5.6。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "openai",
+    baseUrl: OPENAI,
+    model: "gpt-5.6-sol",
+  }),
+  gpt({
+    id: "gpt-5.6-terra",
+    vendor: "OpenAI",
+    label: "GPT-5.6 Terra",
+    summary:
+      "上一代均衡款。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "openai",
+    baseUrl: OPENAI,
+    model: "gpt-5.6-terra",
+  }),
+  gpt({
+    id: "gpt-5.6-luna",
+    vendor: "OpenAI",
+    label: "GPT-5.6 Luna",
+    summary:
+      "上一代高性价比。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "openai",
+    baseUrl: OPENAI,
+    model: "gpt-5.6-luna",
+  }),
+  {
+    id: "qwen3.8-max",
+    vendor: "通义千问",
+    label: "Qwen3.8 Max",
+    summary:
+      "旗舰，快照 qwen3.8-max-0902。思考默认开启（xhigh），可关闭；支持图片、视频。1M 内同一价格，没有分档。",
+    provider: "qwen",
+    baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
+    model: "qwen3.8-max",
+    ...limits(1000000, 131072),
+    ...shared,
+    vision: true,
+    thinkingStyle: "qwen-effort",
+    tokenField: "max_completion_tokens",
+    omitSampling: true,
+    reasoningEfforts: ["none", "low", "medium", "xhigh"],
+    reasoningEffort: "xhigh",
+    temperature: 0.6,
+    topP: 0.95,
+  },
+  {
+    id: "qwen3.8-flash",
+    vendor: "通义千问",
+    label: "Qwen3.8 Flash",
+    summary:
+      "轻量。思考默认开启（xhigh），可关闭；支持图片、视频。1M 内同一价格，没有分档。",
+    provider: "qwen",
+    baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
+    model: "qwen3.8-flash",
+    ...limits(1000000, 131072),
+    ...shared,
+    vision: true,
+    thinkingStyle: "qwen-effort",
+    tokenField: "max_completion_tokens",
+    omitSampling: true,
+    reasoningEfforts: ["none", "low", "medium", "xhigh"],
+    reasoningEffort: "xhigh",
+    temperature: 0.6,
+    topP: 0.95,
+  },
+  gpt({
+    id: "bedrock-gpt-6-astra",
+    vendor: "亚马逊 Bedrock",
+    label: "GPT-6 Astra",
+    summary:
+      "美国跨区推理。思考不能关闭，预填 medium；支持看图。runtime 不支持结构化输出，已关闭 JSON 输出。输入超过 272K 整单按长上下文计价。",
+    provider: "bedrock",
+    baseUrl: BEDROCK,
+    model: "us.openai.gpt-6-astra",
+    json: false,
+    reasoningEfforts: ASTRA_EFFORTS,
+  }),
+  gpt({
+    id: "bedrock-gpt-6-sol",
+    vendor: "亚马逊 Bedrock",
+    label: "GPT-6 Sol",
+    summary:
+      "AWS 模型卡尚未发布，模型 ID 取自 OpenAI 的 Bedrock 指南，JSON 输出先关闭。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "bedrock",
+    baseUrl: BEDROCK,
+    model: "us.openai.gpt-6-sol",
+    json: false,
+  }),
+  gpt({
+    id: "bedrock-gpt-6-luna",
+    vendor: "亚马逊 Bedrock",
+    label: "GPT-6 Luna",
+    summary:
+      "AWS 模型卡尚未发布，模型 ID 取自 OpenAI 的 Bedrock 指南，JSON 输出先关闭。思考默认 medium，可关闭；支持看图。输入超过 272K 整单按长上下文计价。",
+    provider: "bedrock",
+    baseUrl: BEDROCK,
+    model: "us.openai.gpt-6-luna",
+    json: false,
+  }),
+  ...OPENCODE_GO_MODELS,
+  gpt({
+    id: "opencode-zen-gpt-6-astra",
+    vendor: "OpenCode Zen",
+    label: "GPT-6 Astra",
+    summary: "OpenCode Zen · GPT-6 系列旗舰 · Responses 接口。",
+    provider: "opencode-zen",
+    baseUrl: OPENCODE_ZEN,
+    model: "gpt-6-astra",
+    apiProtocol: "responses",
+    reasoningEfforts: ASTRA_EFFORTS,
+  }),
+  gpt({
+    id: "opencode-zen-gpt-6-sol",
+    vendor: "OpenCode Zen",
+    label: "GPT-6 Sol",
+    summary: "OpenCode Zen · GPT-6 均衡款 · Responses 接口。",
+    provider: "opencode-zen",
+    baseUrl: OPENCODE_ZEN,
+    model: "gpt-6-sol",
+    apiProtocol: "responses",
+  }),
+  gpt({
+    id: "opencode-zen-gpt-6-luna",
+    vendor: "OpenCode Zen",
+    label: "GPT-6 Luna",
+    summary: "OpenCode Zen · GPT-6 轻量款 · Responses 接口。",
+    provider: "opencode-zen",
+    baseUrl: OPENCODE_ZEN,
+    model: "gpt-6-luna",
+    apiProtocol: "responses",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "deepseek-v4.1-flash",
+    label: "DeepSeek V4.1 Flash",
+    limits: limits(1000000, 393216),
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    summary: "OpenCode Zen · 快速推理 · Chat Completions 接口。",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "deepseek-v4-pro",
+    label: "DeepSeek V4 Pro",
+    limits: limits(1000000, 393216),
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    summary: "OpenCode Zen · DeepSeek V4 旗舰 · Chat Completions 接口。",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "deepseek-v4-flash",
+    label: "DeepSeek V4 Flash",
+    limits: limits(1000000, 393216),
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    summary: "OpenCode Zen · 轻量高速 · Chat Completions 接口。",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "glm-5.3-flash",
+    label: "GLM-5.3 Flash",
+    vision: true,
+    limits: limits(1000000, 131072),
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    summary: "OpenCode Zen · 智谱轻量多模态 · Chat Completions 接口。",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "glm-5.3",
+    label: "GLM-5.3",
+    limits: limits(1000000, 131072),
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    summary: "OpenCode Zen · 智谱旗舰 · Chat Completions 接口。",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "kimi-k3",
+    label: "Kimi K3",
+    vision: true,
+    limits: limits(1048576, 131072),
+    reasoningEfforts: ["none", "low", "high", "max"],
+    reasoningEffort: "high",
+    summary: "OpenCode Zen · Kimi 旗舰 · Chat Completions 接口。",
+  }),
+  // Qwen3.8 Max is only served through Go; Zen offers Qwen3.8 Flash.
+  openCodeModel({
+    channel: "zen",
+    model: "qwen3.8-flash",
+    label: "Qwen3.8 Flash",
+    protocol: "anthropic",
+    vision: true,
+    limits: limits(1000000, 131072),
+    summary: "OpenCode Zen · 通义千问轻量多模态 · Anthropic Messages 接口。",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "minimax-m3",
+    label: "MiniMax M3",
+    limits: limits(128000, 8192),
+    reasoningEfforts: ["none", "low", "medium", "high"],
+    summary: "OpenCode Zen · MiniMax 主力模型 · Chat Completions 接口。",
+  }),
+  openCodeModel({
+    channel: "zen",
+    model: "space-bunny-free",
+    label: "Space Bunny Free",
+    limits: {
+      contextWindow: 1000000,
+      maxInputTokens: 128000,
+      maxOutputTokens: 8192,
+    },
+    vision: true,
+    reasoningEfforts: [
+      "none",
+      "minimal",
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+      "max",
+    ],
+    reasoningEffort: "low",
+    summary:
+      "OpenCode Zen · 当前限时免费 · Chat Completions 接口。参考 OpenRouter Space Bunny Alpha 规格：1M 上下文、最高 524K 输出，支持图像/视频、推理、工具和 JSON；其匿名第三方提供方可能保留提示和回答。LuckyTri 默认输入 128K、输出 8K、低推理；当前图片链路可用，视频链路未支持。Zen 路由能力与数据政策可能不同。",
+  }),
+  {
+    id: "custom",
+    vendor: "自定义",
+    label: "自定义兼容接口",
+    summary: "OpenAI 兼容接口。上下文先按 128K 填，思考档可按供应商再改。",
+    provider: "custom",
+    baseUrl: "",
+    model: "",
+    ...limits(128000, 8192),
+    embedding: false,
+    embeddingModel: "",
+    timeoutMs: 90000,
+    system: true,
+    json: true,
+    tools: false,
+    vision: false,
+    thinkingStyle: "openai",
+    tokenField: "max_tokens",
+    reasoningEfforts: [
+      "none",
+      "minimal",
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+      "max",
+    ],
+    reasoningEffort: "none",
+    temperature: 0.85,
+    topP: 1,
+  },
+  {
+    id: "openrouter",
+    vendor: "OpenRouter",
+    label: "OpenRouter · 自选模型",
+    summary:
+      "已填好 OpenRouter API 地址。请自行填写模型 ID、API Key，并按所选模型设置上下文、输出和能力参数。",
+    provider: "openrouter",
+    baseUrl: OPENROUTER,
+    model: "",
+    ...limits(128000, 8192),
+    embedding: false,
+    embeddingModel: "",
+    timeoutMs: 90000,
+    system: true,
+    json: true,
+    tools: false,
+    vision: false,
+    thinkingStyle: "openai",
+    tokenField: "max_tokens",
+    reasoningEfforts: [
+      "none",
+      "minimal",
+      "low",
+      "medium",
+      "high",
+      "xhigh",
+      "max",
+    ],
+    reasoningEffort: "none",
+    temperature: 0.85,
+    topP: 1,
+  },
+];
+
+export const MODEL_PRESETS = {
+  custom: { label: "自定义兼容接口", baseUrl: "", model: "" },
+  deepseek: {
+    label: "DeepSeek",
+    baseUrl: "https://api.deepseek.com",
+    model: "deepseek-flash",
+  },
+  openai: {
+    label: "OpenAI",
+    baseUrl: OPENAI,
+    model: "gpt-6-astra",
+  },
+  bedrock: {
+    label: "亚马逊 Bedrock",
+    baseUrl: BEDROCK,
+    model: "us.openai.gpt-6-astra",
+  },
+  qwen: {
+    label: "通义千问",
+    baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
+    model: "qwen3.8-max",
+  },
+  moonshot: {
+    label: "Kimi",
+    baseUrl: "https://api.moonshot.cn/v1",
+    model: "kimi-k3",
+  },
+  kimi: {
+    label: "Kimi",
+    baseUrl: "https://api.moonshot.cn/v1",
+    model: "kimi-k3",
+  },
+  zhipu: {
+    label: "智谱 GLM",
+    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
+    model: "glm-5.3",
+  },
+  glm: {
+    label: "智谱 GLM",
+    baseUrl: "https://open.bigmodel.cn/api/paas/v4",
+    model: "glm-5.3",
+  },
+  mimo: {
+    label: "小米 MiMo",
+    baseUrl: "https://api.xiaomimimo.com/v1",
+    model: "mimo-v2.6-pro",
+  },
+  siliconflow: {
+    label: "SiliconFlow",
+    baseUrl: "https://api.siliconflow.cn/v1",
+    model: "",
+  },
+  openrouter: {
+    label: "OpenRouter",
+    baseUrl: OPENROUTER,
+    model: "",
+  },
+  "volcengine-coding-plan": {
+    label: "火山方舟 Coding Plan",
+    baseUrl: VOLCENGINE_CODING_PLAN,
+    model: "ark-code-latest",
+  },
+};
+
+export const REASONING_EFFORTS = [
+  { id: "none", label: "关闭" },
+  { id: "minimal", label: "极低" },
+  { id: "low", label: "低" },
+  { id: "medium", label: "中" },
+  { id: "high", label: "高" },
+  { id: "xhigh", label: "极高" },
+  { id: "max", label: "最深" },
+];