@lobehub/chat 1.111.10 → 1.111.11

package/packages/model-runtime/src/utils/modelParse.ts

@@ -1,5 +1,7 @@
 import type { ChatModelCard } from '@/types/llm';
 
+import type { ModelProviderKey } from '../types';
+
 export interface ModelProcessorConfig {
   excludeKeywords?: readonly string[]; // 对符合的模型不添加标签
   functionCallKeywords?: readonly string[];
@@ -26,7 +28,12 @@ export const MODEL_LIST_CONFIGS = {
   llama: {
     functionCallKeywords: ['llama-3.2', 'llama-3.3', 'llama-4'],
     reasoningKeywords: [],
-    visionKeywords: [],
+    visionKeywords: ['llava'],
+  },
+  moonshot: {
+    functionCallKeywords: ['moonshot', 'kimi'],
+    reasoningKeywords: ['thinking'],
+    visionKeywords: ['vision', 'kimi-latest', 'kimi-thinking-preview'],
   },
   openai: {
     excludeKeywords: ['audio'],
@@ -45,7 +52,7 @@ export const MODEL_LIST_CONFIGS = {
       'qwen2.5',
       'qwen3',
     ],
-    reasoningKeywords: ['qvq', 'qwq', 'qwen3'],
+    reasoningKeywords: ['qvq', 'qwq', 'qwen3', '!-instruct-', '!-coder-'],
     visionKeywords: ['qvq', 'vl'],
   },
   v0: {
@@ -54,9 +61,14 @@ export const MODEL_LIST_CONFIGS = {
     visionKeywords: ['v0'],
   },
   volcengine: {
-    functionCallKeywords: ['doubao-1.5'],
-    reasoningKeywords: ['thinking', '-r1'],
-    visionKeywords: ['vision', '-m'],
+    functionCallKeywords: ['1.5', '1-5', '1.6', '1-6'],
+    reasoningKeywords: ['thinking', 'seed', 'ui-tars'],
+    visionKeywords: ['vision', '-m', 'seed', 'ui-tars'],
+  },
+  xai: {
+    functionCallKeywords: ['grok'],
+    reasoningKeywords: ['mini', 'grok-4'],
+    visionKeywords: ['vision', 'grok-4'],
   },
   zeroone: {
     functionCallKeywords: ['fc'],
@@ -65,7 +77,7 @@ export const MODEL_LIST_CONFIGS = {
   zhipu: {
     functionCallKeywords: ['glm-4', 'glm-z1'],
     reasoningKeywords: ['glm-zero', 'glm-z1', 'glm-4.5'],
-    visionKeywords: ['glm-4v'],
+    visionKeywords: ['glm-4v', 'glm-4.1v', 'glm-4.5v'],
   },
 } as const;
 
@@ -74,15 +86,107 @@ export const PROVIDER_DETECTION_CONFIG = {
   anthropic: ['claude'],
   deepseek: ['deepseek'],
   google: ['gemini'],
-  llama: ['llama'],
+  llama: ['llama', 'llava'],
+  moonshot: ['moonshot', 'kimi'],
   openai: ['o1', 'o3', 'o4', 'gpt-'],
   qwen: ['qwen', 'qwq', 'qvq'],
   v0: ['v0'],
   volcengine: ['doubao'],
+  xai: ['grok'],
   zeroone: ['yi-'],
   zhipu: ['glm'],
 } as const;
 
+// 图像模型关键词配置
+export const IMAGE_MODEL_KEYWORDS = [
+  'dall-e',
+  'dalle',
+  'midjourney',
+  'stable-diffusion',
+  'sd',
+  'flux',
+  'imagen',
+  'firefly',
+  'cogview',
+  'wanxiang',
+  'DESCRIBE',
+  'UPSCALE',
+  '-image',
+  '^V3',
+  '^V_2',
+  '^V_1',
+] as const;
+
+/**
+ * 检测关键词列表是否匹配模型ID（支持多种匹配模式）
+ * @param modelId 模型ID（小写）
+ * @param keywords 关键词列表，支持以下前缀：
+ *   - ^ 开头：只在模型ID开头匹配
+ *   - ! 开头：排除匹配，优先级最高
+ *   - 无前缀：包含匹配（默认行为）
+ * @returns 是否匹配（排除逻辑优先）
+ */
+const isKeywordListMatch = (modelId: string, keywords: readonly string[]): boolean => {
+  // 先检查排除规则（感叹号开头）
+  const excludeKeywords = keywords.filter((keyword) => keyword.startsWith('!'));
+  const includeKeywords = keywords.filter((keyword) => !keyword.startsWith('!'));
+
+  // 如果匹配任何排除规则，直接返回 false
+  for (const excludeKeyword of excludeKeywords) {
+    const keywordWithoutPrefix = excludeKeyword.slice(1);
+    const isMatch = keywordWithoutPrefix.startsWith('^')
+      ? modelId.startsWith(keywordWithoutPrefix.slice(1))
+      : modelId.includes(keywordWithoutPrefix);
+
+    if (isMatch) {
+      return false;
+    }
+  }
+
+  // 检查包含规则
+  return includeKeywords.some((keyword) => {
+    if (keyword.startsWith('^')) {
+      // ^ 开头则只在开头匹配
+      const keywordWithoutPrefix = keyword.slice(1);
+      return modelId.startsWith(keywordWithoutPrefix);
+    }
+    // 默认行为：包含匹配
+    return modelId.includes(keyword);
+  });
+};
+
+/**
+ * 根据提供商类型查找对应的本地模型配置
+ * @param modelId 模型ID
+ * @param provider 提供商类型
+ * @returns 匹配的本地模型配置
+ */
+const findKnownModelByProvider = async (
+  modelId: string,
+  provider: keyof typeof MODEL_LIST_CONFIGS,
+): Promise<any> => {
+  const lowerModelId = modelId.toLowerCase();
+
+  try {
+    // 动态构建导入路径
+    const modulePath = `@/config/aiModels/${provider}`;
+
+    // 尝试动态导入对应的配置文件
+    const moduleImport = await import(modulePath);
+    const providerModels = moduleImport.default;
+
+    // 如果导入成功且有数据，进行查找
+    if (Array.isArray(providerModels)) {
+      return providerModels.find((m) => m.id.toLowerCase() === lowerModelId);
+    }
+
+    return null;
+  } catch {
+    // 如果导入失败（文件不存在或其他错误），返回 null
+    return null;
+  }
+};
+
 /**
  * 检测单个模型的提供商类型
  * @param modelId 模型ID
@@ -92,7 +196,7 @@ export const detectModelProvider = (modelId: string): keyof typeof MODEL_LIST_CO
   const lowerModelId = modelId.toLowerCase();
 
   for (const [provider, keywords] of Object.entries(PROVIDER_DETECTION_CONFIG)) {
-    const hasKeyword = keywords.some((keyword) => lowerModelId.includes(keyword));
+    const hasKeyword = isKeywordListMatch(lowerModelId, keywords);
 
     if (hasKeyword && provider in MODEL_LIST_CONFIGS) {
       return provider as keyof typeof MODEL_LIST_CONFIGS;
@@ -102,6 +206,44 @@ export const detectModelProvider = (modelId: string): keyof typeof MODEL_LIST_CO
   return 'openai';
 };
 
+/**
+ * 将时间戳转换为日期字符串
+ * @param timestamp 时间戳（秒）
+ * @returns 格式化的日期字符串 (YYYY-MM-DD)
+ */
+const formatTimestampToDate = (timestamp: number): string => {
+  const date = new Date(timestamp * 1000); // 将秒转换为毫秒
+  return date.toISOString().split('T')[0]; // 返回 YYYY-MM-DD 格式
+};
+
+/**
+ * 处理 releasedAt 字段
+ * @param model 模型对象
+ * @param knownModel 已知模型配置
+ * @returns 处理后的 releasedAt 值
+ */
+const processReleasedAt = (model: any, knownModel?: any): string | undefined => {
+  // 优先检查 model.created
+  if (model.created !== undefined && model.created !== null) {
+    // 检查是否为时间戳格式
+    if (typeof model.created === 'number' && model.created > 1_630_000_000) {
+      // AiHubMix 错误时间戳为 1626777600
+      return formatTimestampToDate(model.created);
+    }
+    // 如果 created 是字符串且已经是日期格式，直接返回
+    if (typeof model.created === 'string') {
+      // Anthropic：若为 '2025-02-19T00:00:00Z' 只取日期部分
+      if (/^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}Z$/.test(model.created)) {
+        return model.created.split('T')[0];
+      }
+      return model.created;
+    }
+  }
+
+  // 回退到原有逻辑
+  return model.releasedAt ?? knownModel?.releasedAt ?? undefined;
+};
+
 /**
  * 处理模型卡片的通用逻辑
  */
@@ -117,31 +259,41 @@ const processModelCard = (
     excludeKeywords = [],
   } = config;
 
-  const isExcludedModel = excludeKeywords.some((keyword) =>
-    model.id.toLowerCase().includes(keyword),
-  );
+  const isExcludedModel = isKeywordListMatch(model.id.toLowerCase(), excludeKeywords);
 
   return {
     contextWindowTokens: model.contextWindowTokens ?? knownModel?.contextWindowTokens ?? undefined,
-    displayName: model.displayName ?? knownModel?.displayName ?? model.id,
-    enabled: knownModel?.enabled || false,
+    description: model.description ?? knownModel?.description ?? '',
+    displayName: (model.displayName ?? knownModel?.displayName ?? model.id)
+      .replaceAll(/\s*[(（][^)）]*[)）]\s*/g, '')
+      .trim(), // 去除括号内容
+    enabled: model?.enabled || false,
     functionCall:
-      (functionCallKeywords.some((keyword) => model.id.toLowerCase().includes(keyword)) &&
-        !isExcludedModel) ||
-      knownModel?.abilities?.functionCall ||
-      false,
+      model.functionCall ??
+      knownModel?.abilities?.functionCall ??
+      ((isKeywordListMatch(model.id.toLowerCase(), functionCallKeywords) && !isExcludedModel) ||
+        false),
     id: model.id,
     maxOutput: model.maxOutput ?? knownModel?.maxOutput ?? undefined,
+    // pricing: knownModel?.pricing ?? undefined,
     reasoning:
-      reasoningKeywords.some((keyword) => model.id.toLowerCase().includes(keyword)) ||
-      knownModel?.abilities?.reasoning ||
-      false,
-    type: model.type || knownModel?.type || 'chat',
+      model.reasoning ??
+      knownModel?.abilities?.reasoning ??
+      (isKeywordListMatch(model.id.toLowerCase(), reasoningKeywords) || false),
+    releasedAt: processReleasedAt(model, knownModel),
+    type:
+      model.type ||
+      knownModel?.type ||
+      (isKeywordListMatch(
+        model.id.toLowerCase(),
+        IMAGE_MODEL_KEYWORDS.map((k) => k.toLowerCase()),
+      )
+        ? 'image'
+        : 'chat'),
     vision:
-      (visionKeywords.some((keyword) => model.id.toLowerCase().includes(keyword)) &&
-        !isExcludedModel) ||
-      knownModel?.abilities?.vision ||
-      false,
+      model.vision ??
+      knownModel?.abilities?.vision ??
+      ((isKeywordListMatch(model.id.toLowerCase(), visionKeywords) && !isExcludedModel) || false),
   };
 };
 
@@ -149,45 +301,91 @@ const processModelCard = (
  * 处理单一提供商的模型列表
  * @param modelList 模型列表
  * @param config 提供商配置
+ * @param provider 提供商类型（可选，用于优先匹配对应的本地配置）
  * @returns 处理后的模型卡片列表
  */
 export const processModelList = async (
   modelList: Array<{ id: string }>,
   config: ModelProcessorConfig,
+  provider?: keyof typeof MODEL_LIST_CONFIGS,
 ): Promise<ChatModelCard[]> => {
   const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
 
-  return modelList
-    .map((model) => {
-      const knownModel = LOBE_DEFAULT_MODEL_LIST.find(
-        (m) => model.id.toLowerCase() === m.id.toLowerCase(),
-      );
+  return Promise.all(
+    modelList.map(async (model) => {
+      let knownModel: any = null;
+
+      // 如果提供了provider，优先使用提供商特定的配置
+      if (provider) {
+        knownModel = await findKnownModelByProvider(model.id, provider);
+      }
+
+      // 如果未找到，回退到全局配置
+      if (!knownModel) {
+        knownModel = LOBE_DEFAULT_MODEL_LIST.find(
+          (m) => model.id.toLowerCase() === m.id.toLowerCase(),
+        );
+      }
 
       return processModelCard(model, config, knownModel);
-    })
-    .filter(Boolean);
+    }),
+  ).then((results) => results.filter(Boolean));
 };
 
 /**
  * 处理混合提供商的模型列表
  * @param modelList 模型列表
+ * @param providerid 可选的提供商ID，用于获取其本地配置文件
  * @returns 处理后的模型卡片列表
  */
 export const processMultiProviderModelList = async (
   modelList: Array<{ id: string }>,
+  providerid?: ModelProviderKey,
 ): Promise<ChatModelCard[]> => {
   const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
 
-  return modelList
-    .map((model) => {
+  // 如果提供了 providerid，尝试获取该提供商的本地配置
+  let providerLocalConfig: any[] | null = null;
+  if (providerid) {
+    try {
+      const modulePath = `@/config/aiModels/${providerid}`;
+      const moduleImport = await import(modulePath);
+      providerLocalConfig = moduleImport.default;
+    } catch {
+      // 如果配置文件不存在或导入失败，保持为 null
+      providerLocalConfig = null;
+    }
+  }
+
+  return Promise.all(
+    modelList.map(async (model) => {
       const detectedProvider = detectModelProvider(model.id);
       const config = MODEL_LIST_CONFIGS[detectedProvider];
 
-      const knownModel = LOBE_DEFAULT_MODEL_LIST.find(
-        (m) => model.id.toLowerCase() === m.id.toLowerCase(),
-      );
+      // 优先使用提供商特定的配置
+      let knownModel = await findKnownModelByProvider(model.id, detectedProvider);
 
-      return processModelCard(model, config, knownModel);
-    })
-    .filter(Boolean);
+      // 如果未找到，回退到全局配置
+      if (!knownModel) {
+        knownModel = LOBE_DEFAULT_MODEL_LIST.find(
+          (m) => model.id.toLowerCase() === m.id.toLowerCase(),
+        );
+      }
+
+      // 如果提供了 providerid 且有本地配置，尝试从中获取模型的 enabled 状态
+      let providerLocalModelConfig = null;
+      if (providerLocalConfig && Array.isArray(providerLocalConfig)) {
+        providerLocalModelConfig = providerLocalConfig.find((m) => m.id === model.id);
+      }
+
+      const processedModel = processModelCard(model, config, knownModel);
+
+      // 如果找到了本地配置中的模型，使用其 enabled 状态
+      if (providerLocalModelConfig && typeof providerLocalModelConfig.enabled === 'boolean') {
+        processedModel.enabled = providerLocalModelConfig.enabled;
+      }
+
+      return processedModel;
+    }),
+  ).then((results) => results.filter(Boolean));
 };

package/packages/model-runtime/src/utils/openaiCompatibleFactory/index.test.ts

@@ -1392,6 +1392,12 @@ describe('LobeOpenAICompatibleFactory', () => {
           maxOutput: 4096,
           pricing: {
             units: [
+              {
+                name: 'textInput_cacheRead',
+                rate: 0.03,
+                strategy: 'fixed',
+                unit: 'millionTokens',
+              },
               {
                 name: 'textInput',
                 rate: 0.25,
@@ -1404,6 +1410,18 @@ describe('LobeOpenAICompatibleFactory', () => {
                 strategy: 'fixed',
                 unit: 'millionTokens',
               },
+              {
+                lookup: {
+                  prices: {
+                    '1h': 0.5,
+                    '5m': 0.3,
+                  },
+                  pricingParams: ['ttl'],
+                },
+                name: 'textInput_cacheWrite',
+                strategy: 'lookup',
+                unit: 'millionTokens',
+              },
             ],
           },
           providerId: 'anthropic',

package/packages/model-runtime/src/utils/streams/openai/openai.ts

@@ -188,6 +188,18 @@ const transformOpenAIStream = (
       let reasoning_content = (() => {
         if ('reasoning_content' in item.delta) return item.delta.reasoning_content;
         if ('reasoning' in item.delta) return item.delta.reasoning;
+        // Handle content array format with thinking blocks (e.g. mistral AI Magistral model)
+        if ('content' in item.delta && Array.isArray(item.delta.content)) {
+          return item.delta.content
+            .filter((block: any) => block.type === 'thinking' && Array.isArray(block.thinking))
+            .map((block: any) => 
+              block.thinking
+                .filter((thinkItem: any) => thinkItem.type === 'text' && thinkItem.text)
+                .map((thinkItem: any) => thinkItem.text)
+                .join('')
+            )
+            .join('');
+        }
         return null;
       })();
 

package/packages/model-runtime/src/v0/index.ts

@@ -1,5 +1,5 @@
 import { ModelProvider } from '../types';
-import { processMultiProviderModelList } from '../utils/modelParse';
+import { MODEL_LIST_CONFIGS, processModelList } from '../utils/modelParse';
 import { createOpenAICompatibleRuntime } from '../utils/openaiCompatibleFactory';
 
 export interface V0ModelCard {
@@ -15,7 +15,7 @@ export const LobeV0AI = createOpenAICompatibleRuntime({
     const modelsPage = (await client.models.list()) as any;
     const modelList: V0ModelCard[] = modelsPage.data;
 
-    return processMultiProviderModelList(modelList);
+    return processModelList(modelList, MODEL_LIST_CONFIGS.v0, 'v0');
   },
   provider: ModelProvider.V0,
 });

package/packages/model-runtime/src/volcengine/index.ts

@@ -37,7 +37,7 @@ export const LobeVolcengineAI = createOpenAICompatibleRuntime({
     const modelsPage = (await client.models.list()) as any;
     const modelList: VolcengineModelCard[] = modelsPage.data;
 
-    return processModelList(modelList, MODEL_LIST_CONFIGS.volcengine);
+    return processModelList(modelList, MODEL_LIST_CONFIGS.volcengine, 'volcengine');
   },
   provider: ModelProvider.Volcengine,
 });

package/packages/model-runtime/src/xai/index.ts

@@ -1,6 +1,5 @@
-import type { ChatModelCard } from '@/types/llm';
-
 import { ModelProvider } from '../types';
+import { MODEL_LIST_CONFIGS, processModelList } from '../utils/modelParse';
 import { createOpenAICompatibleRuntime } from '../utils/openaiCompatibleFactory';
 
 export interface XAIModelCard {
@@ -55,31 +54,10 @@ export const LobeXAI = createOpenAICompatibleRuntime({
     chatCompletion: () => process.env.DEBUG_XAI_CHAT_COMPLETION === '1',
   },
   models: async ({ client }) => {
-    const { LOBE_DEFAULT_MODEL_LIST } = await import('@/config/aiModels');
-
     const modelsPage = (await client.models.list()) as any;
     const modelList: XAIModelCard[] = modelsPage.data;
 
-    return modelList
-      .map((model) => {
-        const knownModel = LOBE_DEFAULT_MODEL_LIST.find(
-          (m) => model.id.toLowerCase() === m.id.toLowerCase(),
-        );
-
-        return {
-          contextWindowTokens: knownModel?.contextWindowTokens ?? undefined,
-          displayName: knownModel?.displayName ?? undefined,
-          enabled: knownModel?.enabled || false,
-          functionCall: knownModel?.abilities?.functionCall || false,
-          id: model.id,
-          reasoning: knownModel?.abilities?.reasoning || false,
-          vision:
-            model.id.toLowerCase().includes('vision') ||
-            knownModel?.abilities?.functionCall ||
-            false,
-        };
-      })
-      .filter(Boolean) as ChatModelCard[];
+    return processModelList(modelList, MODEL_LIST_CONFIGS.xai, 'xai');
   },
   provider: ModelProvider.XAI,
 });

package/packages/model-runtime/src/zhipu/index.ts

@@ -134,7 +134,7 @@ export const LobeZhipuAI = createOpenAICompatibleRuntime({
       displayName: model.modelName,
       id: model.modelCode,
     }));
-    return processModelList(standardModelList, MODEL_LIST_CONFIGS.zhipu);
+    return processModelList(standardModelList, MODEL_LIST_CONFIGS.zhipu, 'zhipu');
   },
   provider: ModelProvider.ZhiPu,
 });

package/src/config/aiModels/aihubmix.ts

@@ -328,7 +328,6 @@ const aihubmixModels: AIChatModelCard[] = [
   {
     abilities: {
       functionCall: true,
-      search: true,
     },
     contextWindowTokens: 131_072,
     description:
@@ -344,16 +343,12 @@ const aihubmixModels: AIChatModelCard[] = [
       ],
     },
     releasedAt: '2025-04-03',
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
       reasoning: true,
-      search: true,
     },
     contextWindowTokens: 131_072,
     description:
@@ -370,7 +365,6 @@ const aihubmixModels: AIChatModelCard[] = [
     releasedAt: '2025-04-03',
     settings: {
       extendParams: ['reasoningEffort'],
-      searchImpl: 'params',
     },
     type: 'chat',
   },
@@ -780,4 +774,4 @@ const aihubmixModels: AIChatModelCard[] = [
   },
 ];
 
-export default aihubmixModels;
+export default aihubmixModels;

package/src/config/aiModels/anthropic.ts

@@ -113,7 +113,6 @@ const anthropicChatModels: AIChatModelCard[] = [
     description:
       'Claude 3.7 Sonnet 是 Anthropic 迄今为止最智能的模型，也是市场上首个混合推理模型。Claude 3.7 Sonnet 可以产生近乎即时的响应或延长的逐步思考，用户可以清晰地看到这些过程。Sonnet 特别擅长编程、数据科学、视觉处理、代理任务。',
     displayName: 'Claude 3.7 Sonnet',
-    enabled: true,
     id: 'claude-3-7-sonnet-20250219',
     maxOutput: 64_000,
     pricing: {
@@ -213,9 +212,9 @@ const anthropicChatModels: AIChatModelCard[] = [
     maxOutput: 8192,
     pricing: {
       units: [
-        { name: 'textInput_cacheRead', rate: 0.1, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textInput', rate: 1, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 5, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput_cacheRead', rate: 0.08, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput', rate: 0.8, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 4, strategy: 'fixed', unit: 'millionTokens' },
         {
           lookup: { prices: { '1h': 1.6, '5m': 1 }, pricingParams: ['ttl'] },
           name: 'textInput_cacheWrite',
@@ -243,8 +242,15 @@ const anthropicChatModels: AIChatModelCard[] = [
     maxOutput: 4096,
     pricing: {
       units: [
+        { name: 'textInput_cacheRead', rate: 0.03, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textInput', rate: 0.25, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textOutput', rate: 1.25, strategy: 'fixed', unit: 'millionTokens' },
+        {
+          lookup: { prices: { '1h': 0.5, '5m': 0.3 }, pricingParams: ['ttl'] },
+          name: 'textInput_cacheWrite',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
       ],
     },
     releasedAt: '2024-03-07',
@@ -266,8 +272,15 @@ const anthropicChatModels: AIChatModelCard[] = [
     maxOutput: 4096,
     pricing: {
       units: [
+        { name: 'textInput_cacheRead', rate: 0.3, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textInput', rate: 3, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textOutput', rate: 15, strategy: 'fixed', unit: 'millionTokens' },
+        {
+          lookup: { prices: { '1h': 6, '5m': 3.75 }, pricingParams: ['ttl'] },
+          name: 'textInput_cacheWrite',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
       ],
     },
     releasedAt: '2024-02-29',
@@ -286,8 +299,15 @@ const anthropicChatModels: AIChatModelCard[] = [
     maxOutput: 4096,
     pricing: {
       units: [
+        { name: 'textInput_cacheRead', rate: 1.5, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textInput', rate: 15, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textOutput', rate: 75, strategy: 'fixed', unit: 'millionTokens' },
+        {
+          lookup: { prices: { '1h': 30, '5m': 18.75 }, pricingParams: ['ttl'] },
+          name: 'textInput_cacheWrite',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
       ],
     },
     releasedAt: '2024-02-29',