@henjicc/ai-sdk 0.8.0 → 0.10.0

package/dist/llm/modelCatalogEntries.js

@@ -12,9 +12,27 @@
  * 否则用户会得到"勾了但发过去没反应"的静默失效。
  */
 export const LLM_MODEL_CATALOG_ENTRIES = [
+    {
+        id: 'deepseek-v4.1-flash',
+        displayName: 'DeepSeek V4.1 Flash',
+        vendor: 'DeepSeek',
+        // 方舟写作 deepseek-v4-1-flash-<快照>（含经济版），百炼另有四位快照 -1017。
+        aliases: ['deepseek-v4-1-flash', 'deepseek-v4-1-flash-economy', 'deepseek-v4.1-flash-1017'],
+        input: { image: true, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        reasoning: true,
+        sampling: true,
+        contextWindow: 1_000_000,
+        maxOutputTokens: 384_000,
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        note: '能力按 DeepSeek 官方 V4.1 Flash（与原厂 deepseek-flash 一致）：图片理解、函数工具、JSON、思考，上下文 1,000,000、最大输出 384,000。派欧云与方舟走 Chat；百炼官方列入 Responses 支持清单。各平台价格见对应供应商资料。',
+        docs: 'docs/model-adaptation/供应商/派欧云.md',
+    },
     // 硅基流动模型页与 Chat VLM 契约；未确认的视频字段和输出上限不猜测。
     ...[
-        { id: 'kimi-k2.7-code', displayName: 'Kimi K2.7 Code', vendor: 'Moonshot' },
+        { id: 'kimi-k2.7-code', displayName: 'Kimi K2.7 Code', vendor: 'Moonshot', deprecated: 'Kimi 于 2026 年底下线全部 K2 系列，10-31 起不再对新用户开放；请改用 K3 / K3.1 Pro' },
         { id: 'qwen3.8-27b', displayName: 'Qwen3.8-27B', vendor: 'Qwen' },
     ].map(model => ({
         ...model,
@@ -45,6 +63,22 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         note: 'Groq Production 模型；reasoning_effort 仅支持 low / medium / high，且不支持 reasoning_format。',
         docs: 'docs/model-adaptation/GPT-OSS-20B/GPT-OSS-20B_Groq.md',
     },
+    {
+        id: 'gpt-oss-120b',
+        displayName: 'GPT-OSS 120B',
+        vendor: 'OpenAI',
+        input: { image: false, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        structuredOutputWithReasoning: true,
+        reasoning: true,
+        sampling: true,
+        contextWindow: 131_072,
+        maxOutputTokens: 65_536,
+        note: 'Groq Production 模型，参数约定与 20B 相同；reasoning_effort 仅支持 low / medium / high，且不支持 reasoning_format。',
+        docs: 'docs/model-adaptation/供应商/Groq.md',
+    },
     // ---------------- DeepSeek ----------------
     {
         id: 'deepseek-flash',
@@ -66,6 +100,7 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         id: 'deepseek-v4-flash',
         displayName: 'DeepSeek V4 Flash',
         vendor: 'DeepSeek',
+        aliases: ['deepseek-v4-flash-ga', 'deepseek-v4-flash-0731'],
         input: { image: false, video: false, audio: false },
         toolCall: true,
         parallelTools: false,
@@ -82,6 +117,7 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         id: 'deepseek-v4-pro',
         displayName: 'DeepSeek V4 Pro',
         vendor: 'DeepSeek',
+        aliases: ['deepseek-v4-pro-ga', 'deepseek-v4-pro-0813'],
         input: { image: false, video: false, audio: false },
         toolCall: true,
         parallelTools: false,
@@ -93,22 +129,6 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         apiProtocols: ['openai-responses', 'openai-compatible'],
         docs: 'docs/llm-adaptation/供应商/DeepSeek.md',
     },
-    {
-        id: 'deepseek-v4-flash-vision-exp',
-        displayName: 'DeepSeek V4 Flash Vision (实验)',
-        vendor: 'DeepSeek',
-        input: { image: true, video: false, audio: false },
-        toolCall: true,
-        parallelTools: false,
-        structuredOutputMode: 'json',
-        reasoning: true,
-        sampling: true,
-        contextWindow: 1_000_000,
-        maxOutputTokens: 384_000,
-        apiProtocols: ['openai-responses', 'openai-compatible'],
-        note: '图片输入只在 Responses API 下可用；SDK 会在直连 DeepSeek 时自动选择 Responses。',
-        docs: 'docs/llm-adaptation/供应商/DeepSeek.md',
-    },
     // ---------------- Kimi（Moonshot） ----------------
     {
         id: 'kimi-k3',
@@ -128,6 +148,40 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         note: '思考模式始终开启无法关闭；图片必须用 base64 data URL，不接受公网直链；视频要先经 Files API 上传再用 ms://<file-id> 引用。',
         docs: 'docs/llm-adaptation/供应商/Kimi.md',
     },
+    {
+        id: 'kimi-k3.1-pro',
+        displayName: 'Kimi K3.1 Pro',
+        vendor: 'Moonshot',
+        aliases: ['kimi-k3-1-pro'],
+        // 方舟把它列在视觉理解能力里，但没写明是否收视频，视频保守不开。
+        input: { image: true, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        reasoning: true,
+        // 原厂文档写明 K3 系列采样参数固定、必须省略。
+        sampling: false,
+        contextWindow: 1_048_576,
+        maxOutputTokens: null,
+        note: '经方舟与百炼托管调用：思考默认开启，可关闭；百炼标注结构化输出为 JSON Mode，内置工具只在 Responses 下可用（项目走 Chat）。原厂直连的能力见 Kimi 供应商预设。',
+        docs: 'docs/llm-adaptation/供应商/火山引擎.md',
+    },
+    {
+        id: 'kimi-k3.1-pro-highspeed',
+        displayName: 'Kimi K3.1 Pro Highspeed',
+        vendor: 'Moonshot',
+        input: { image: true, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'schema',
+        structuredOutputWithReasoning: true,
+        reasoning: true,
+        sampling: false,
+        contextWindow: 1_048_576,
+        maxOutputTokens: 131_072,
+        note: 'K3.1 Pro 的高速版，速度约 2～3 倍，只在 Kimi 原厂提供；思考无法关闭。',
+        docs: 'docs/llm-adaptation/供应商/Kimi.md',
+    },
     {
         id: 'kimi-k2.6',
         displayName: 'Kimi K2.6',
@@ -141,6 +195,7 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         contextWindow: null,
         maxOutputTokens: null,
         note: '沿用项目既有内置配置的模态标注，未逐项核对官方文档；上下文与输出上限留空由动态验证或用户补齐。',
+        deprecated: 'Kimi 于 2026 年底下线全部 K2 系列，10-31 起不再对新用户开放；请改用 K3 / K3.1 Pro',
         docs: 'docs/llm-adaptation/供应商/Kimi.md',
     },
     // ---------------- 智谱 GLM ----------------
@@ -149,6 +204,8 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         { id: 'glm-5.3-flashx', displayName: 'GLM-5.3-FlashX' },
     ].map(model => ({
         ...model,
+        // 方舟托管写作 glm-5-3-flash-<快照>。
+        ...(model.id === 'glm-5.3-flash' ? { aliases: ['glm-5-3-flash'] } : {}),
         vendor: '智谱',
         input: { image: true, video: true, audio: false, file: true },
         toolCall: true,
@@ -179,6 +236,24 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         note: '始终开启思考且无法关闭；官方明确只支持文本模态，需要看图看视频请改用 GLM-5V-Turbo。',
         docs: 'docs/llm-adaptation/供应商/智谱GLM.md',
     },
+    {
+        id: 'glm-5.2',
+        displayName: 'GLM-5.2',
+        vendor: '智谱',
+        aliases: ['glm-5-2'],
+        input: { image: false, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        // 方舟与百炼都没有给 GLM-5.2 标结构化输出，保守不开。
+        structuredOutputMode: 'none',
+        reasoning: true,
+        sampling: true,
+        contextWindow: 1_048_576,
+        maxOutputTokens: 131_072,
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        note: '经方舟与百炼托管调用；纯文本。百炼列入 Responses 支持清单。',
+        docs: 'docs/llm-adaptation/供应商/火山引擎.md',
+    },
     {
         id: 'glm-5v-turbo',
         displayName: 'GLM-5V-Turbo',
@@ -221,8 +296,9 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         structuredOutputMode: 'json',
         reasoning: true,
         sampling: true,
-        contextWindow: 262_144,
+        contextWindow: 1_048_576,
         maxOutputTokens: 262_144,
+        snapshots: { 'doubao-seed-2-1-pro-260628': { contextWindow: 262_144 } },
         apiProtocols: ['openai-responses', 'openai-compatible'],
         docs: 'docs/llm-adaptation/供应商/火山引擎.md',
     },
@@ -241,6 +317,58 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         apiProtocols: ['openai-responses', 'openai-compatible'],
         docs: 'docs/llm-adaptation/供应商/火山引擎.md',
     },
+    {
+        id: 'doubao-seed-2-1-lite',
+        displayName: 'Doubao-Seed-2.1-Lite',
+        vendor: '火山引擎',
+        // 官方另支持音频理解，但项目还没核对方舟的音频内容块写法，先不开。
+        input: { image: true, video: true, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        reasoning: true,
+        sampling: true,
+        contextWindow: 1_048_576,
+        maxOutputTokens: 262_144,
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        docs: 'docs/llm-adaptation/供应商/火山引擎.md',
+    },
+    ...[
+        { id: 'doubao-seed-2-0-lite', displayName: 'Doubao-Seed-2.0-Lite' },
+        { id: 'doubao-seed-2-0-mini', displayName: 'Doubao-Seed-2.0-Mini' },
+    ].map(model => ({
+        ...model,
+        vendor: '火山引擎',
+        input: { image: true, video: true, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        reasoning: true,
+        sampling: true,
+        // 上下文 256K，单次最大输入 224K。
+        contextWindow: 262_144,
+        maxOutputTokens: 131_072,
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        note: '长期支持版（LTS）快照 260428；260215 快照即将下线。官方另支持音频理解，项目暂未开放。',
+        docs: 'docs/llm-adaptation/供应商/火山引擎.md',
+    })),
+    {
+        id: 'doubao-seed-character',
+        displayName: 'Doubao-Seed-Character',
+        vendor: '火山引擎',
+        input: { image: true, video: true, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        reasoning: true,
+        sampling: true,
+        contextWindow: 131_072,
+        maxOutputTokens: 32_768,
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        note: '角色扮演模型；能力按 260628 快照，旧快照 251128 不支持思考与多模态。',
+        snapshots: { 'doubao-seed-character-251128': { input: { image: false, video: false, audio: false }, reasoning: false, structuredOutputMode: 'none' } },
+        docs: 'docs/llm-adaptation/供应商/火山引擎.md',
+    },
     // ---------------- 小米 MiMo ----------------
     ...[
         { id: 'mimo-v2.6-pro', displayName: 'MiMo-V2.6-Pro' },
@@ -261,11 +389,13 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         note: '支持 Chat Completions 与 Responses；思考只有开关，无强度分级。UltraSpeed 需要联系供应商开通。',
         docs: 'docs/llm-adaptation/供应商/小米MiMo.md',
     })),
+    // ---------------- 阿里云百炼 Qwen ----------------
     {
-        id: 'mimo-v2.5-pro',
-        displayName: 'MiMo-V2.5-Pro',
-        vendor: '小米',
-        input: { image: false, video: false, audio: false },
+        id: 'qwen3.8-max',
+        displayName: 'Qwen3.8-Max',
+        vendor: '阿里云百炼',
+        aliases: ['qwen3.8-max-prime', 'qwen3.8-max-0902'],
+        input: { image: true, video: true, audio: false },
         toolCall: true,
         parallelTools: false,
         structuredOutputMode: 'json',
@@ -273,14 +403,17 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         sampling: true,
         contextWindow: 1_000_000,
         maxOutputTokens: 131_072,
-        note: '多轮工具调用时官方建议把历史轮次的 reasoning_content 一并回传。',
-        docs: 'docs/llm-adaptation/供应商/小米MiMo.md',
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        note: '必须走多模态接口：即使是纯文字对话，content 也要写成内容块数组，直接传字符串会报 url error。',
+        docs: 'docs/llm-adaptation/供应商/百炼Qwen.md',
     },
-    {
-        id: 'mimo-v2.5',
-        displayName: 'MiMo-V2.5',
-        vendor: '小米',
-        input: { image: true, video: true, audio: true },
+    ...[
+        { id: 'qwen3.8-flash', displayName: 'Qwen3.8-Flash' },
+        { id: 'qwen3.7-plus', displayName: 'Qwen3.7-Plus' },
+    ].map(model => ({
+        ...model,
+        vendor: '阿里云百炼',
+        input: { image: true, video: true, audio: false },
         toolCall: true,
         parallelTools: false,
         structuredOutputMode: 'json',
@@ -288,29 +421,42 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         sampling: true,
         contextWindow: 1_000_000,
         maxOutputTokens: 131_072,
-        note: '官方标注全模态理解，图片 / 音频 / 视频均可作为输入。',
-        docs: 'docs/llm-adaptation/供应商/小米MiMo.md',
-    },
-    // ---------------- 阿里云百炼 Qwen ----------------
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        docs: 'docs/llm-adaptation/供应商/百炼Qwen.md',
+    })),
     {
-        id: 'qwen3.8-max',
-        displayName: 'Qwen3.8-Max',
-        vendor: '阿里云百炼',
-        aliases: ['qwen3.8-max-prime'],
-        input: { image: true, video: true, audio: false },
+        id: 'step-5-preview',
+        displayName: 'Step-5-Preview',
+        vendor: '阶跃星辰',
+        // 只在百炼三方直供核对过：未写明输入模态，按纯文本处理。
+        input: { image: false, video: false, audio: false },
         toolCall: true,
         parallelTools: false,
         structuredOutputMode: 'json',
         reasoning: true,
         sampling: true,
-        // 官方文档只写"超长文档与长视频"，没有给出可引用的上下文 token 数，留空走回退预算。
-        contextWindow: null,
+        contextWindow: 1_000_000,
         maxOutputTokens: null,
-        apiProtocols: ['openai-responses', 'openai-compatible'],
-        note: '必须走多模态接口：即使是纯文字对话，content 也要写成内容块数组，直接传字符串会报 url error。',
         docs: 'docs/llm-adaptation/供应商/百炼Qwen.md',
     },
     // ---------------- MiniMax ----------------
+    {
+        id: 'minimax-m3.1-flash',
+        displayName: 'MiniMax-M3.1-Flash',
+        vendor: 'MiniMax',
+        // 百炼资料没写输入模态，按纯文本处理。
+        input: { image: false, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        // 百炼标注不支持结构化输出、仅思考模式。
+        structuredOutputMode: 'none',
+        reasoning: true,
+        sampling: true,
+        contextWindow: 1_000_000,
+        maxOutputTokens: null,
+        note: '经百炼三方直供调用，只有思考模式；不支持结构化输出与内置工具。',
+        docs: 'docs/llm-adaptation/供应商/百炼Qwen.md',
+    },
     {
         id: 'minimax-m3',
         displayName: 'MiniMax-M3',
@@ -328,3 +474,12 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         docs: 'docs/llm-adaptation/供应商/MiniMax.md',
     },
 ];
+/** 原厂旧名已转到新 Flash；网关同名仍保留上方的独立旧契约。 */
+export const DEEPSEEK_OFFICIAL_FLASH_ALIASES = [
+    'deepseek-v4-flash', 'deepseek-v4-flash-vision-exp',
+].map(id => ({
+    ...LLM_MODEL_CATALOG_ENTRIES.find(entry => entry.id === 'deepseek-flash'),
+    id,
+    displayName: 'DeepSeek V4.1 Flash（原厂旧别名）',
+    note: '原厂旧实体已下线，旧名称重定向到 deepseek-flash；支持图片及 Chat / Responses API。网关同名条目不使用此覆盖。',
+}));

package/dist/llm/providerPresets.d.ts

@@ -1,4 +1,5 @@
 import type { LlmApiProtocol } from './providerProtocol.js';
+import { type TokenListPricing } from '../pricing/tokenPricing.js';
 import type { LlmCapabilities, LlmModelConfig, LlmProviderConfig, LlmReasoningEffort } from './types.js';
 /**
  * 内置供应商预设。
@@ -29,6 +30,8 @@ export interface LlmProviderPreset {
     modelCapabilities?: Readonly<Record<string, Partial<LlmCapabilities>>>;
     /** 这家供应商已确认可走 Responses 的具体模型；未列出的模型继续走 Chat Completions。 */
     responsesModelIds?: readonly string[];
+    /** 这家供应商对推荐模型公布的按量标准价，出处写在 docs 指向的资料里；没查到的模型不登记。 */
+    modelPricing?: Readonly<Record<string, TokenListPricing>>;
     websiteUrl: string;
     apiKeyUrl: string;
     /** 该预设的资料出处，仓库内相对路径 */

package/dist/llm/providerPresets.js

@@ -4,6 +4,7 @@ import { GROQ_PROVIDER_PRESET } from './groq/preset.js';
 import { SILICONFLOW_PROVIDER_PRESET } from './siliconflow/preset.js';
 import { createBigmodelProvider } from './bigmodel/preset.js';
 import { BIGMODEL_ENDPOINT_PROFILE_FAMILY } from './bigmodel/profiles.js';
+import { tokenListPrice, tokenPrice } from '../pricing/tokenPricing.js';
 import { findProviderMetadata } from '../providers/metadata.js';
 function providerMetadata(providerId) {
     const metadata = findProviderMetadata(providerId);
@@ -11,6 +12,7 @@ function providerMetadata(providerId) {
         throw new Error(`[provider_metadata_missing] provider "${providerId}" is not registered`);
     return metadata;
 }
+const DEEPSEEK_PEAK_HOURS = '北京时间工作日 09:00–12:00、14:00–18:00';
 export const LLM_PROVIDER_PRESETS = [
     GROQ_PROVIDER_PRESET,
     SILICONFLOW_PROVIDER_PRESET,
@@ -22,12 +24,29 @@ export const LLM_PROVIDER_PRESETS = [
         baseUrl: DEFAULT_PPIO_BASE_URL,
         reasoning: { enabled: false, effort: 'high' },
         reasoningConfigurable: false,
+        // 2026-11-07 按派欧云模型列表核对；已过时的条目在建模型时自动跳过。
         modelIds: [
             'deepseek/deepseek-v4-pro',
-            'deepseek/deepseek-v4-flash',
+            'deepseek/deepseek-v4.1-flash',
+            'moonshotai/kimi-k3.1-pro',
+            'moonshotai/kimi-k3',
+            'zai-org/glm-5.3',
+            'xiaomimimo/mimo-v2.6-pro',
+            'minimax/minimax-m3.1-flash',
             'moonshotai/kimi-k2.6',
-            'xiaomimimo/mimo-v2.5-pro',
         ],
+        // 模型详情页刊例价；页面只显示当前时段的实际价，DeepSeek V4.1 Flash 的高峰价在“查看其他时段刊例价”里。
+        // Kimi 另计缓存写入（K3.1 Pro 5 分钟 13、K3 5 分钟 20 元/百万），不在这里登记。
+        modelPricing: {
+            'deepseek/deepseek-v4-pro': tokenListPrice('CNY', 9, 27, 3),
+            'deepseek/deepseek-v4.1-flash': { ...tokenListPrice('CNY', 2, 8, 0.04), offPeak: { ...tokenPrice(1, 4, 0.02), peakHours: DEEPSEEK_PEAK_HOURS } },
+            'moonshotai/kimi-k3.1-pro': tokenListPrice('CNY', 13, 68, 1.3),
+            'moonshotai/kimi-k3': tokenListPrice('CNY', 20, 100, 2),
+            'zai-org/glm-5.3': tokenListPrice('CNY', 8, 28, 2),
+            'xiaomimimo/mimo-v2.6-pro': tokenListPrice('CNY', 3, 6, 0.025),
+            'minimax/minimax-m3.1-flash': tokenListPrice('CNY', 1, 4, 0.1),
+            'moonshotai/kimi-k2.6': tokenListPrice('CNY', 6.5, 27, 1.1),
+        },
         websiteUrl: providerMetadata(DEFAULT_PPIO_PROVIDER_ID).websiteUrl,
         apiKeyUrl: providerMetadata(DEFAULT_PPIO_PROVIDER_ID).apiKeyUrl,
         docs: 'docs/llm-adaptation/README.md',
@@ -41,8 +60,13 @@ export const LLM_PROVIDER_PRESETS = [
         baseUrl: DEFAULT_DEEPSEEK_BASE_URL,
         reasoning: { enabled: true, effort: 'high' },
         reasoningConfigurable: true,
-        modelIds: ['deepseek-flash'],
-        responsesModelIds: ['deepseek-flash', 'deepseek-v4-pro', 'deepseek-v4-flash', 'deepseek-v4-flash-vision-exp'],
+        modelIds: ['deepseek-flash', 'deepseek-v4-pro'],
+        responsesModelIds: ['deepseek-flash', 'deepseek-v4-pro'],
+        // 官方价格页按北京时间工作日高峰 / 其余闲时两档计价（高峰不含法定节假日）。
+        modelPricing: {
+            'deepseek-flash': { ...tokenListPrice('CNY', 2, 8, 0.04), offPeak: { ...tokenPrice(1, 4, 0.02), peakHours: DEEPSEEK_PEAK_HOURS } },
+            'deepseek-v4-pro': { ...tokenListPrice('CNY', 9, 27, 0.3), offPeak: { ...tokenPrice(4.5, 13.5, 0.15), peakHours: DEEPSEEK_PEAK_HOURS } },
+        },
         websiteUrl: providerMetadata(DEFAULT_DEEPSEEK_PROVIDER_ID).websiteUrl,
         apiKeyUrl: providerMetadata(DEFAULT_DEEPSEEK_PROVIDER_ID).apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/DeepSeek.md',
@@ -55,8 +79,34 @@ export const LLM_PROVIDER_PRESETS = [
         baseUrl: 'https://ark.cn-beijing.volces.com/api/v3',
         reasoning: { enabled: true, effort: 'high' },
         reasoningConfigurable: true,
-        modelIds: ['doubao-seed-evolving', 'doubao-seed-2-1-pro-260628', 'doubao-seed-2-1-turbo-260628'],
-        responsesModelIds: ['doubao-seed-evolving', 'doubao-seed-2-1-pro-260628', 'doubao-seed-2-1-turbo-260628'],
+        modelIds: [
+            'doubao-seed-evolving', 'doubao-seed-2-1-pro-260915', 'doubao-seed-2-1-lite-260915', 'doubao-seed-2-1-turbo-260628',
+            'doubao-seed-2-0-lite-260428', 'doubao-seed-2-0-mini-260428', 'doubao-seed-character-260628',
+            'deepseek-v4-1-flash-261017', 'deepseek-v4-1-flash-economy-261017', 'deepseek-v4-pro-ga-260813', 'deepseek-v4-flash-ga-260731',
+            'kimi-k3-1-pro-261024', 'glm-5-3-flash-260828', 'glm-5-2-260617',
+        ],
+        // 方舟说明 250615 以后的大模型默认支持 Responses；托管的第三方模型未逐个核实字段兼容，留在 Chat。
+        responsesModelIds: [
+            'doubao-seed-evolving', 'doubao-seed-2-1-pro-260915', 'doubao-seed-2-1-lite-260915', 'doubao-seed-2-1-turbo-260628',
+            'doubao-seed-2-0-lite-260428', 'doubao-seed-2-0-mini-260428', 'doubao-seed-character-260628',
+        ],
+        // 常规在线推理价；分档按单次请求输入长度落档（官方单位为千 token）。音频输入另有单价，不在这里展开。
+        modelPricing: {
+            'doubao-seed-evolving': tokenListPrice('CNY', 6, 30, 1.2),
+            'doubao-seed-2-1-pro-260915': tokenListPrice('CNY', 6, 30, 1.2),
+            'doubao-seed-2-1-lite-260915': tokenListPrice('CNY', 0.8, 2.7, 0.16),
+            'doubao-seed-2-1-turbo-260628': tokenListPrice('CNY', 3, 15, 0.6),
+            'doubao-seed-2-0-lite-260428': { ...tokenListPrice('CNY', 0.6, 3.6, 0.12), tiers: [{ ...tokenPrice(0.9, 5.4, 0.18), aboveInputTokens: 32_000 }, { ...tokenPrice(1.8, 10.8, 0.36), aboveInputTokens: 128_000 }] },
+            'doubao-seed-2-0-mini-260428': { ...tokenListPrice('CNY', 0.2, 2, 0.04), tiers: [{ ...tokenPrice(0.4, 4, 0.08), aboveInputTokens: 32_000 }, { ...tokenPrice(0.8, 8, 0.16), aboveInputTokens: 128_000 }] },
+            'doubao-seed-character-260628': { ...tokenListPrice('CNY', 0.8, 2, 0.16), tiers: [{ ...tokenPrice(1.2, 6, 0.16), aboveInputTokens: 32_000 }] },
+            'deepseek-v4-1-flash-261017': { ...tokenListPrice('CNY', 2, 8, 0.04), offPeak: { ...tokenPrice(1, 4, 0.02), peakHours: DEEPSEEK_PEAK_HOURS } },
+            'deepseek-v4-1-flash-economy-261017': { ...tokenListPrice('CNY', 1.6, 6.4, 0.032), offPeak: { ...tokenPrice(0.8, 3.2, 0.016), peakHours: DEEPSEEK_PEAK_HOURS } },
+            'deepseek-v4-pro-ga-260813': tokenListPrice('CNY', 9, 27, 0.3),
+            'deepseek-v4-flash-ga-260731': tokenListPrice('CNY', 3, 9, 0.1),
+            'kimi-k3-1-pro-261024': tokenListPrice('CNY', 13, 68, 1.3),
+            'glm-5-3-flash-260828': tokenListPrice('CNY', 0.8, 2.8, 0.23),
+            'glm-5-2-260617': tokenListPrice('CNY', 8, 28, 2),
+        },
         websiteUrl: providerMetadata('volcengine').websiteUrl,
         apiKeyUrl: providerMetadata('volcengine').apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/火山引擎.md',
@@ -68,14 +118,27 @@ export const LLM_PROVIDER_PRESETS = [
         adapter: 'openai',
         apiProtocol: 'openai-compatible',
         baseUrl: 'https://api.moonshot.cn/v1',
-        // K3 始终思考、无法关闭，只能调强度，所以不给"关闭"这个选项。
+        // K3 系列始终思考、无法关闭，只能调强度，所以不给"关闭"这个选项。
         reasoning: { enabled: true, effort: 'max' },
         reasoningConfigurable: true,
-        modelIds: ['kimi-k3'],
+        modelIds: ['kimi-k3.1-pro', 'kimi-k3.1-pro-highspeed', 'kimi-k3'],
+        // 原厂 K3 系列：采样参数固定、必须省略；结构化输出支持 json_schema 且可与思考同用。
+        // 视频只收 Files API 上传后的 ms://<file-id>，SDK 没有这条上传链路，所以不声明视频。
+        modelCapabilities: Object.fromEntries(['kimi-k3.1-pro', 'kimi-k3.1-pro-highspeed', 'kimi-k3'].map(modelId => [modelId, {
+                image: true, video: false, audio: false,
+                structuredOutputMode: 'schema', structuredOutputWithReasoning: true, sampling: false,
+                contextWindow: 1_048_576, maxOutputTokens: 131_072,
+            }])),
+        // 缓存写入按存放时长另计（5 分钟与输入同价、1 小时翻倍），不在这里登记；不按上下文长度分档。
+        modelPricing: {
+            'kimi-k3.1-pro': tokenListPrice('CNY', 13, 68, 1.3),
+            'kimi-k3.1-pro-highspeed': tokenListPrice('CNY', 26, 136, 1.3),
+            'kimi-k3': tokenListPrice('CNY', 20, 100, 2),
+        },
         websiteUrl: providerMetadata('kimi').websiteUrl,
         apiKeyUrl: providerMetadata('kimi').apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/Kimi.md',
-        note: '文档站是 platform.kimi.com，但接口域名仍是 moonshot.cn，不要混用；K3 需要充值后才能调用。K3 的思考无法关闭，思考模式选「关闭」时按最低强度发送。',
+        note: '文档站是 platform.kimi.com，但接口域名仍是 moonshot.cn，不要混用；K3 系列需要充值后才能调用。K3.1 Pro Highspeed 速度约为 K3.1 Pro 的 2～3 倍。K3 系列思考无法关闭，思考模式选「关闭」时按最低强度发送。',
     },
     {
         providerId: 'bigmodel',
@@ -87,6 +150,13 @@ export const LLM_PROVIDER_PRESETS = [
         reasoningConfigurable: true,
         modelIds: ['glm-5.3', 'glm-5v-turbo', 'glm-5.3-flash', 'glm-5.3-flashx'],
         responsesModelIds: ['glm-5.3'],
+        // 中国大陆标准价（划线前价格，不含限时折扣）。
+        modelPricing: {
+            'glm-5.3': tokenListPrice('CNY', 8, 28, 2),
+            'glm-5v-turbo': { ...tokenListPrice('CNY', 5, 22, 1.2), tiers: [{ ...tokenPrice(7, 26, 1.8), aboveInputTokens: 32_000 }] },
+            'glm-5.3-flash': tokenListPrice('CNY', 0.8, 2.8, 0.23),
+            'glm-5.3-flashx': tokenListPrice('CNY', 2, 7, 0.57),
+        },
         websiteUrl: providerMetadata('bigmodel').websiteUrl,
         apiKeyUrl: providerMetadata('bigmodel').apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/智谱GLM.md',
@@ -102,6 +172,10 @@ export const LLM_PROVIDER_PRESETS = [
         reasoningConfigurable: true,
         modelIds: ['mimo-v2.6-pro', 'mimo-v2.6-flash'],
         responsesModelIds: ['mimo-v2.6-pro', 'mimo-v2.6-flash', 'mimo-v2.6-pro-ultraspeed'],
+        modelPricing: {
+            'mimo-v2.6-pro': tokenListPrice('CNY', 3, 6, 0.025),
+            'mimo-v2.6-flash': tokenListPrice('CNY', 1, 2, 0.02),
+        },
         websiteUrl: providerMetadata('mimo').websiteUrl,
         apiKeyUrl: providerMetadata('mimo').apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/小米MiMo.md',
@@ -112,11 +186,25 @@ export const LLM_PROVIDER_PRESETS = [
         displayName: 'MiniMax',
         adapter: 'openai',
         apiProtocol: 'openai-compatible',
-        baseUrl: 'https://api.minimaxi.com/v1',
+        // 官方国内接口域名已改为 minimax.cn（旧域名 minimaxi.com 仍可用）。
+        baseUrl: 'https://api.minimax.cn/v1',
         reasoning: { enabled: true, effort: 'high' },
         reasoningConfigurable: true,
-        modelIds: ['MiniMax-M3'],
+        modelIds: ['MiniMax-M3.1-Flash', 'MiniMax-M3'],
         responsesModelIds: ['MiniMax-M3'],
+        // 原厂 M3.1 Flash 收图片与视频；思考无法关闭；temperature / top_p 被固定，不下发；
+        // 官方没写结构化输出与最大输出长度，保持未声明。
+        modelCapabilities: {
+            'MiniMax-M3.1-Flash': {
+                image: true, video: true, audio: false,
+                structuredOutputMode: 'none', sampling: false, contextWindow: 1_000_000, maxOutputTokens: null,
+            },
+        },
+        // M3 按输入长度两档；官方标注“永久五折”，按折后价登记。
+        modelPricing: {
+            'MiniMax-M3.1-Flash': tokenListPrice('CNY', 1, 4, 0.1),
+            'MiniMax-M3': { ...tokenListPrice('CNY', 2.1, 8.4, 0.42), tiers: [{ ...tokenPrice(4.2, 16.8, 0.84), aboveInputTokens: 512_000 }] },
+        },
         websiteUrl: providerMetadata('minimax').websiteUrl,
         apiKeyUrl: providerMetadata('minimax').apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/MiniMax.md',
@@ -131,12 +219,31 @@ export const LLM_PROVIDER_PRESETS = [
         baseUrlHint: '形如 https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1，在百炼控制台模型页的「查看代码」里复制当前值。',
         reasoning: { enabled: true, effort: 'high' },
         reasoningConfigurable: true,
-        modelIds: ['qwen3.8-max'],
-        responsesModelIds: ['qwen3.8-max'],
+        modelIds: [
+            'qwen3.8-max', 'qwen3.8-flash', 'qwen3.7-plus',
+            'deepseek-v4.1-flash', 'deepseek-v4-pro-0813', 'kimi-k3.1-pro', 'kimi-k3', 'glm-5.3',
+            'MiniMax/MiniMax-M3.1-Flash', 'xiaomi/mimo-v2.6-pro-ultraspeed', 'stepfun/step-5-preview',
+        ],
+        // 百炼 Responses 支持清单里的模型；kimi-k3 虽在清单里，但原厂 Kimi 不支持 Responses，目录按模型登记协议，留在 Chat。
+        responsesModelIds: ['qwen3.8-max', 'qwen3.8-flash', 'qwen3.7-plus', 'deepseek-v4.1-flash', 'deepseek-v4-pro-0813', 'glm-5.3'],
+        // 华北2（北京）原价；DeepSeek 按忙时（北京时间 08:00–22:00）/ 闲时两档。计费页没有单列缓存价的不登记。
+        modelPricing: {
+            'qwen3.8-max': tokenListPrice('CNY', 12, 36),
+            'qwen3.8-flash': tokenListPrice('CNY', 0.8, 2.7),
+            'qwen3.7-plus': { ...tokenListPrice('CNY', 2, 8), tiers: [{ ...tokenPrice(6, 24), aboveInputTokens: 256_000 }] },
+            'deepseek-v4.1-flash': { ...tokenListPrice('CNY', 2, 8), offPeak: { ...tokenPrice(1, 4), peakHours: '北京时间 08:00–22:00' } },
+            'deepseek-v4-pro-0813': { ...tokenListPrice('CNY', 9, 27), offPeak: { ...tokenPrice(4.5, 13.5), peakHours: '北京时间 08:00–22:00' } },
+            'kimi-k3.1-pro': tokenListPrice('CNY', 13, 68, 1.3),
+            'kimi-k3': tokenListPrice('CNY', 20, 100),
+            'glm-5.3': tokenListPrice('CNY', 8, 28),
+            'MiniMax/MiniMax-M3.1-Flash': tokenListPrice('CNY', 1, 4),
+            'xiaomi/mimo-v2.6-pro-ultraspeed': tokenListPrice('CNY', 30, 60),
+            'stepfun/step-5-preview': tokenListPrice('CNY', 7, 20),
+        },
         websiteUrl: providerMetadata('bailian').websiteUrl,
         apiKeyUrl: providerMetadata('bailian').apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/百炼Qwen.md',
-        note: 'qwen3.8-max 必须走多模态消息格式，纯文本也要写成内容块数组。',
+        note: 'qwen3.8-max 走 Chat 时必须用多模态消息格式，纯文本也要写成内容块数组。三方直供模型的 Model ID 带厂商前缀（如 MiniMax/MiniMax-M3.1-Flash）。',
     },
 ];
 export function findLlmProviderPreset(providerId) {
@@ -147,9 +254,13 @@ export function resolvePresetModelApiProtocol(preset, modelId, endpointProfile)
     if (preset.providerId === 'bigmodel' && (endpointProfile ?? 'cn') !== 'cn') {
         return 'openai-compatible';
     }
-    const modelEntry = findLlmModelCatalogEntry(modelId);
+    const modelEntry = findLlmModelCatalogEntry(modelId, preset.providerId);
     const normalizedModelId = modelEntry?.id ?? modelId.trim().toLowerCase();
-    const supportsResponses = preset.responsesModelIds?.some(candidate => ((findLlmModelCatalogEntry(candidate)?.id ?? candidate.trim().toLowerCase()) === normalizedModelId)) === true;
+    // 带厂商前缀的是平台上的另一条渠道（如百炼 ZHIPU/GLM-5.3 与 glm-5.3），Responses 支持要按原写法单独确认。
+    const routed = (value) => value.includes('/');
+    const supportsResponses = preset.responsesModelIds?.some(candidate => (routed(candidate) || routed(modelId)
+        ? candidate.trim().toLowerCase() === modelId.trim().toLowerCase()
+        : (findLlmModelCatalogEntry(candidate, preset.providerId)?.id ?? candidate.trim().toLowerCase()) === normalizedModelId)) === true;
     return supportsResponses && modelEntry && modelSupportsLlmApiProtocol(modelEntry, 'openai-responses')
         ? 'openai-responses'
         : 'openai-compatible';
@@ -192,21 +303,23 @@ export function createModelsFromPreset(preset, provider) {
     if (!modelIds) {
         throw new Error(`[llm_endpoint_profile_unknown] bigmodel endpoint profile "${provider.endpointProfile}" is unavailable`);
     }
-    return modelIds.map((modelId) => {
-        const entry = findLlmModelCatalogEntry(modelId);
-        return {
-            providerId: provider.providerId,
-            providerFamilyId: provider.providerFamilyId,
-            endpointProfile: provider.endpointProfile,
-            credentialId: provider.credentialId,
-            modelId,
-            displayName: entry?.displayName ?? modelId,
-            adapter: provider.adapter,
-            apiProtocol: resolvePresetModelApiProtocol(preset, modelId, provider.endpointProfile),
-            baseUrl: provider.baseUrl,
-            capabilities: { ...createLlmCapabilitiesForModel(modelId), ...preset.modelCapabilities?.[modelId] },
-            catalogId: entry?.id,
-            enabled: true,
-        };
+    return modelIds.flatMap((modelId) => {
+        const entry = findLlmModelCatalogEntry(modelId, preset.providerId);
+        if (entry?.deprecated)
+            return [];
+        return [{
+                providerId: provider.providerId,
+                providerFamilyId: provider.providerFamilyId,
+                endpointProfile: provider.endpointProfile,
+                credentialId: provider.credentialId,
+                modelId,
+                displayName: entry?.displayName ?? modelId,
+                adapter: provider.adapter,
+                apiProtocol: resolvePresetModelApiProtocol(preset, modelId, provider.endpointProfile),
+                baseUrl: provider.baseUrl,
+                capabilities: { ...createLlmCapabilitiesForModel(modelId, preset.providerId), ...preset.modelCapabilities?.[modelId] },
+                catalogId: entry?.id,
+                enabled: true,
+            }];
     });
 }