@henjicc/ai-sdk 0.9.0 → 0.10.0

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,
+            }];
     });
 }

package/dist/llm/providerReasoningRequest.js

@@ -47,11 +47,19 @@ const PROVIDER_REASONING_BODY = {
     deepseek: reasoning => (reasoning.enabled
         ? { thinking: { type: 'enabled' }, reasoning_effort: pickEffort(reasoning.effort, ['high', 'max']) }
         : { thinking: { type: 'disabled' } }),
-    // Kimi K3：始终思考、官方没有关闭开关，只有 low / high / max 三档。
+    // Kimi K3 系列（K3、K3.1 Pro 与 Highspeed）：始终思考、官方没有关闭开关，只有 low / high / max 三档。
     // 用户选「关闭」时退到最低档，而不是假装关掉——这一点在供应商预设的说明里也写了。
-    kimi: reasoning => ({
-        reasoning_effort: reasoning.enabled ? pickEffort(reasoning.effort, ['low', 'high', 'max']) : 'low',
-    }),
+    kimi: (reasoning, body) => {
+        const model = typeof body.model === 'string' ? body.model.toLowerCase() : '';
+        if (model === 'kimi-k3' || model.startsWith('kimi-k3.1-pro')) {
+            return { reasoning_effort: reasoning.enabled ? pickEffort(reasoning.effort, ['low', 'high', 'max']) : 'low' };
+        }
+        if (model === 'kimi-k2.6')
+            return { thinking: { type: reasoning.enabled ? 'enabled' : 'disabled' } };
+        if (model === 'kimi-k2.7-code')
+            return { thinking: { type: 'enabled', keep: 'all' } };
+        return {};
+    },
     // 智谱 GLM：GLM-5.3 传 thinking.type='disabled' 会直接请求失败（官方迁移说明写明），
     // 而 GLM-5V-Turbo 不传即为关闭。所以关闭时整段不发，让两个模型各自走默认，
     // 而不是发一个会让旗舰模型 400 的字段。
@@ -66,16 +74,25 @@ const PROVIDER_REASONING_BODY = {
             reasoning_effort: pickEffort(reasoning.effort, ['low', 'medium', 'high', 'xhigh', 'max']),
         }
         : { thinking: { type: 'disabled' } }),
+    // MiniMax M3.1 Flash：reasoning_effort 有 low / medium / high / xhigh / max 五档，省略即 max，
+    // 发 none 会 400，所以「关闭」退到最低档。M3 走 Responses，保持通用写法。
+    minimax: (reasoning, body) => {
+        const model = typeof body.model === 'string' ? body.model.toLowerCase() : '';
+        if (model !== 'minimax-m3.1-flash')
+            return defaultReasoningBody(reasoning, body);
+        return { reasoning_effort: reasoning.enabled ? pickEffort(reasoning.effort, ['low', 'medium', 'high', 'xhigh', 'max']) : 'low' };
+    },
     // 阿里云百炼：布尔开关，没有强度分级。
     bailian: reasoning => ({ enable_thinking: reasoning.enabled }),
     // Groq GPT-OSS：只接受 low / medium / high；不支持 reasoning_format。
     // include_reasoning 是独立的返回控制开关，关闭时不再发送 reasoning_effort。
-    groq: reasoning => (reasoning.enabled
-        ? {
-            reasoning_effort: pickEffort(reasoning.effort, ['low', 'medium', 'high']),
-            include_reasoning: true,
-        }
-        : { include_reasoning: false }),
+    // GPT-OSS 不能关闭思考，关闭时只隐藏思考内容；Qwen3.8-27B 收 reasoning_effort: none，能真正关掉。
+    groq: (reasoning, body) => {
+        const model = typeof body.model === 'string' ? body.model.toLowerCase() : '';
+        if (reasoning.enabled)
+            return { reasoning_effort: pickEffort(reasoning.effort, ['low', 'medium', 'high']), include_reasoning: true };
+        return model === 'qwen/qwen3.8-27b' ? { reasoning_effort: 'none' } : { include_reasoning: false };
+    },
 };
 /**
  * 未登记供应商的兜底：只发 OpenAI 事实标准字段，关闭时什么都不发。
@@ -128,7 +145,7 @@ export function applyProviderReasoningRequestBody(providerId, adapter, body, rea
             ? { ...clean, reasoning: { effort: reasoning.enabled ? 'high' : 'none' } }
             : { ...clean, ...build(reasoning, clean) };
     }
-    if (key === 'siliconflow') {
+    if (key === 'siliconflow' || key === 'kimi') {
         const { thinking: _thinking, reasoning_effort: _effort, ...clean } = body;
         return { ...clean, ...build(reasoning, clean) };
     }

package/dist/llm/siliconflow/index.js

@@ -15,7 +15,7 @@ export function createSiliconflowChatRequest(request) {
         adapter: 'openai',
         baseUrl: request.baseUrl?.trim() || SILICONFLOW_BASE_URL,
         capabilities: request.capabilities ?? (known ? {
-            ...createLlmCapabilitiesForModel(modelId),
+            ...createLlmCapabilitiesForModel(modelId, 'siliconflow'),
             ...SILICONFLOW_PROVIDER_PRESET.modelCapabilities?.[modelId],
         } : undefined),
         reasoning: request.reasoning ?? (known ? { ...SILICONFLOW_PROVIDER_PRESET.reasoning } : undefined),

package/dist/llm/siliconflow/preset.js

@@ -1,7 +1,9 @@
 import { findProviderMetadata } from '../../providers/metadata.js';
+import { tokenListPrice, tokenPrice } from '../../pricing/tokenPricing.js';
 export const SILICONFLOW_PROVIDER_ID = 'siliconflow';
 export const SILICONFLOW_BASE_URL = 'https://api.siliconflow.cn/v1';
 export const SILICONFLOW_DEFAULT_MODEL_ID = 'deepseek-ai/DeepSeek-V4-Flash';
+const SILICONFLOW_PEAK_HOURS = '北京时间 00:00–02:00、08:00–24:00';
 const metadata = findProviderMetadata(SILICONFLOW_PROVIDER_ID);
 if (!metadata)
     throw new Error('[provider_metadata_missing] siliconflow');
@@ -14,17 +16,45 @@ export const SILICONFLOW_PROVIDER_PRESET = {
     baseUrl: SILICONFLOW_BASE_URL,
     reasoning: { enabled: true, effort: 'high' },
     reasoningConfigurable: true,
+    // 2026-11-07 按账号模型列表核对；已过时的条目在建模型时自动跳过。
     modelIds: [
         SILICONFLOW_DEFAULT_MODEL_ID,
+        'deepseek-ai/DeepSeek-V4.1-Flash',
+        'deepseek-ai/DeepSeek-V4-Pro',
         'zai-org/GLM-5.3',
+        'zai-org/GLM-5.3-Flash',
+        'zai-org/GLM-5.2',
         'moonshotai/Kimi-K2.7-Code',
         'Qwen/Qwen3.8-27B',
     ],
+    // 托管能力以硅基流动为准（模型广场的结构化字段：vlm、functionCallSupport、jsonModeSupport、contextLen），
+    // 不继承原厂；平台没有单独的输出上限（max_tokens 只受上下文窗口约束），留空。视频输入未取得托管接口的字段契约，一律不声明。
     modelCapabilities: {
         'zai-org/GLM-5.3': {
             image: false, video: false, audio: false, file: false,
             sampling: true, contextWindow: 1_048_576, maxOutputTokens: null,
         },
+        ...Object.fromEntries([
+            ['deepseek-ai/DeepSeek-V4.1-Flash', true],
+            ['deepseek-ai/DeepSeek-V4-Pro', false],
+            ['zai-org/GLM-5.3-Flash', true],
+            ['zai-org/GLM-5.2', false],
+        ].map(([modelId, image]) => [modelId, {
+                image, video: false, audio: false, file: false,
+                toolCall: true, jsonOutput: true, structuredOutputMode: 'json', reasoning: true,
+                contextWindow: 1_048_576, maxOutputTokens: null,
+            }])),
+    },
+    // 控制台模型详情“在线推理”价（单元格按千 tokens 标，这里换算成百万）；DeepSeek Flash 系列每天 02:00–08:00 为闲时价。
+    modelPricing: {
+        [SILICONFLOW_DEFAULT_MODEL_ID]: { ...tokenListPrice('CNY', 3, 9, 0.3), offPeak: { ...tokenPrice(1.5, 4.5, 0.15), peakHours: SILICONFLOW_PEAK_HOURS } },
+        'deepseek-ai/DeepSeek-V4.1-Flash': { ...tokenListPrice('CNY', 2, 8, 0.04), offPeak: { ...tokenPrice(1, 4, 0.02), peakHours: SILICONFLOW_PEAK_HOURS } },
+        'deepseek-ai/DeepSeek-V4-Pro': tokenListPrice('CNY', 12, 24, 1),
+        'zai-org/GLM-5.3': tokenListPrice('CNY', 8, 28, 2),
+        'zai-org/GLM-5.3-Flash': tokenListPrice('CNY', 0.8, 2.8, 0.23),
+        'zai-org/GLM-5.2': tokenListPrice('CNY', 8, 28, 2),
+        'moonshotai/Kimi-K2.7-Code': tokenListPrice('CNY', 6.5, 27, 1.3),
+        'Qwen/Qwen3.8-27B': tokenListPrice('CNY', 3, 12),
     },
     websiteUrl: metadata.websiteUrl,
     apiKeyUrl: metadata.apiKeyUrl,

package/dist/llm/types.d.ts

@@ -234,6 +234,12 @@ export interface LlmTrace {
     elapsedMs: number;
     inputChars: number;
     outputChars: number;
+    /** 发出请求到第一个字的毫秒数（含思考内容） */
+    firstTokenMs?: number | null;
+    /** 首字之后的输出 token 速度（个/秒） */
+    outputTokensPerSecond?: number | null;
+    inputTokens?: number | null;
+    outputTokens?: number | null;
 }
 export type LlmStreamEvent = {
     type: 'Token';

package/dist/packs/models/{ppio/wan-2.5-preview.d.ts → apimart/flux-3-image.d.ts}

@@ -1,5 +1,5 @@
 /** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
-import model from '../../../catalog/ppio/wan-2.5-preview.model.js';
+import model from '../../../catalog/apimart/flux-3-image.model.js';
 import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
 export { model };
 export declare const provider: GenerationClientProviderRegistration;

package/dist/packs/models/apimart/flux-3-image.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/flux-3-image.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/flux-3-video.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/flux-3-video.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/flux-3-video.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/flux-3-video.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/gemini-omni-1.2-flash.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/gemini-omni-1.2-flash.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/gemini-omni-1.2-flash.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/gemini-omni-1.2-flash.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/{modelscope/qwen-image.d.ts → apimart/gpt-image-2.6.d.ts}

@@ -1,5 +1,5 @@
 /** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
-import model from '../../../catalog/modelscope/qwen-image.model.js';
+import model from '../../../catalog/apimart/gpt-image-2.6.model.js';
 import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
 export { model };
 export declare const provider: GenerationClientProviderRegistration;

package/dist/packs/models/apimart/gpt-image-2.6.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/gpt-image-2.6.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/grok-imagine-video-1.5.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/grok-imagine-video-1.5.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/grok-imagine-video-1.5.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/grok-imagine-video-1.5.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/happyhorse-1.1.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/happyhorse-1.1.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/happyhorse-1.1.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/happyhorse-1.1.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/kling-4.0.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/kling-4.0.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/kling-4.0.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/kling-4.0.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/minimax-h3-max.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/minimax-h3-max.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/minimax-h3-max.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/minimax-h3-max.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/{ppio/minimax-hailuo-2.3.d.ts → apimart/nano-banana-2.1.d.ts}

@@ -1,5 +1,5 @@
 /** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
-import model from '../../../catalog/ppio/minimax-hailuo-2.3.model.js';
+import model from '../../../catalog/apimart/nano-banana-2.1.model.js';
 import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
 export { model };
 export declare const provider: GenerationClientProviderRegistration;

package/dist/packs/models/{ppio/minimax-hailuo-2.3.js → apimart/nano-banana-2.1.js}

@@ -1,7 +1,7 @@
 /** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
-import model from '../../../catalog/ppio/minimax-hailuo-2.3.model.js';
-import * as adapter from '../../../providers/ppio.js';
-import { preprocess } from '../../../upload/provider-preprocessors/ppio.js';
+import model from '../../../catalog/apimart/nano-banana-2.1.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
 export { model };
-export const provider = { id: 'ppio', adapter, preprocess };
+export const provider = { id: 'apimart', adapter, preprocess };
 export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/seedream-5.0-flash.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/seedream-5.0-flash.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/seedream-5.0-flash.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/seedream-5.0-flash.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };

package/dist/packs/models/apimart/veo-3.1.d.ts

@@ -0,0 +1,6 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/veo-3.1.model.js';
+import type { GenerationClientProviderRegistration, GenerationPack } from '../../../generation/core.js';
+export { model };
+export declare const provider: GenerationClientProviderRegistration;
+export declare const pack: GenerationPack;

package/dist/packs/models/apimart/veo-3.1.js

@@ -0,0 +1,7 @@
+/** 由 scripts/generate-catalog-index.cjs 自动生成；`pack` 是完整执行单元，`model` 仅供低层目录用途。 */
+import model from '../../../catalog/apimart/veo-3.1.model.js';
+import * as adapter from '../../../providers/apimart.js';
+import { preprocess } from '../../../upload/provider-preprocessors/apimart.js';
+export { model };
+export const provider = { id: 'apimart', adapter, preprocess };
+export const pack = { models: [model], providers: [provider] };