utilitas 1998.2.42 → 1998.2.44

package/lib/alan.mjs

@@ -54,7 +54,7 @@ const [
     CLAUDE_35_HAIKU, CLOUD_37_SONNET, AUDIO, WAV, CHATGPT_MINI, ATTACHMENTS,
     CHAT, OPENAI_VOICE, MEDIUM, LOW, HIGH, GPT_REASONING_EFFORT, THINK,
     THINK_STR, THINK_END, AZURE, TOOLS_STR, TOOLS_END, TOOLS, TEXT, THINKING,
-    OK, FUNC,
+    OK, FUNC, GPT_45,
 ] = [
         'OPENAI', 'GEMINI', 'CHATGPT', 'OPENAI_EMBEDDING', 'GEMINI_EMEDDING',
         'OPENAI_TRAINING', 'OLLAMA', 'CLAUDE', 'gpt-4o-mini', 'gpt-4o', 'o1',
@@ -67,6 +67,7 @@ const [
         '[ATTACHMENTS]', 'CHAT', 'OPENAI_VOICE', 'medium', 'low', 'high',
         'medium', 'think', '<think>', '</think>', 'AZURE', '<tools>',
         '</tools>', 'tools', 'text', 'thinking', 'OK', 'function',
+        'gpt-4.5-preview',
     ];
 
 const [
@@ -147,9 +148,9 @@ const MODELS = {
         contextWindow: 128000,
         imageCostTokens: 1105,
         maxOutputTokens: 16384,
-        requestLimitsRPM: 5000,
-        tokenLimitsTPD: 40000000,
-        tokenLimitsTPM: 4000000,
+        requestLimitsRPM: 10000,
+        tokenLimitsTPD: 1000000000,
+        tokenLimitsTPM: 10000000,
         trainingData: 'Oct 2023',
         json: true,
         vision: true,
@@ -166,9 +167,9 @@ const MODELS = {
         contextWindow: 128000,
         imageCostTokens: 1105,
         maxOutputTokens: 16384,
-        requestLimitsRPM: 5000,
-        tokenLimitsTPD: 100000000,
-        tokenLimitsTPM: 800000,
+        requestLimitsRPM: 10000,
+        tokenLimitsTPD: 20000000,
+        tokenLimitsTPM: 2000000,
         trainingData: 'Oct 2023',
         json: true,
         vision: true,
@@ -185,9 +186,9 @@ const MODELS = {
         contextWindow: 200000,
         imageCostTokens: 1105,
         maxOutputTokens: 100000,
-        requestLimitsRPM: 500,
-        tokenLimitsTPD: 100000000,
-        tokenLimitsTPM: 800000,
+        requestLimitsRPM: 10000,
+        tokenLimitsTPD: 200000000,
+        tokenLimitsTPM: 2000000,
         trainingData: 'Oct 2023',
         json: true,
         reasoning: true,
@@ -205,9 +206,9 @@ const MODELS = {
         contextWindow: 200000,
         imageCostTokens: 1105,
         maxOutputTokens: 100000,
-        requestLimitsRPM: 5000,
-        tokenLimitsTPD: 40000000,
-        tokenLimitsTPM: 4000000,
+        requestLimitsRPM: 10000,
+        tokenLimitsTPD: 1000000000,
+        tokenLimitsTPM: 10000000,
         trainingData: 'Oct 2023',
         json: true,
         reasoning: true,
@@ -221,6 +222,21 @@ const MODELS = {
         //     wav,
         // ],
     },
+    [GPT_45]: {
+        contextWindow: 128000,
+        imageCostTokens: 1105,
+        maxOutputTokens: 16384,
+        requestLimitsRPM: 10000,
+        tokenLimitsTPD: 100000000,
+        tokenLimitsTPM: 1000000,
+        json: true,
+        vision: true,
+        tools: true,
+        supportedMimeTypes: [
+            png, jpeg, gif, webp,
+        ],
+        trainingData: 'Oct 2023',
+    },
     [GEMINI_20_FLASH]: {
         // https://ai.google.dev/gemini-api/docs/models/gemini
         // https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/send-multimodal-prompts?hl=en#gemini-send-multimodal-samples-pdf-nodejs
@@ -628,8 +644,10 @@ const buildClaudeMessage = (text, options) => {
         }
     });
     return String.isString(text) ? {
-        role: options?.role || user,
-        content: [...attachments, { type: TEXT, text }],
+        role: options?.role || user, content: [...attachments, {
+            type: TEXT, text,
+            cache_control: options?.cache_control ?? { type: 'ephemeral' }
+        }],
     } : text;
 };
 
@@ -655,10 +673,11 @@ const streamResp = async (resp, options) => {
 };
 
 const getInfoEnd = text => Math.max(...[THINK_END, TOOLS_END].map(x => {
-    const keyEnd = text.indexOf(`${x}\n`);
+    const keyEnd = text.indexOf(text.endsWith(x) ? x : `${x}\n`);
     return keyEnd >= 0 ? (keyEnd + x.length) : 0;
 }));
 
+// @todo: escape ``` in think and tools
 const packResp = async (resp, options) => {
     if (options?.raw) { return resp; }
     let [txt, audio, references, simpleText, referencesMarkdown, end, json] = [
@@ -1018,18 +1037,25 @@ const promptOllama = async (content, options = {}) => {
 
 const promptClaude = async (content, options = {}) => {
     options.model = options.model || DEFAULT_MODELS[CLAUDE];
-    let [_MODEL, event, text, thinking, signature, result, thinkEnd, tool_use]
-        = [MODELS[options.model], null, '', '', '', options.result ?? '', '', []];
+    let [
+        _MODEL, event, text, thinking, signature, result, thinkEnd, tool_use,
+        responded
+    ] = [
+            MODELS[options.model], null, '', '', '', options.result ?? '', '',
+            [], false
+        ];
     const { client } = await getClaudeClient(options);
     const { history }
         = await buildPrompts(_MODEL, content, { ...options, flavor: CLAUDE });
-    const resp = await client.messages.create({
+    const resp = await client.beta.messages.create({
         model: options.model, max_tokens: _MODEL.maxOutputTokens, ...history,
         stream: true, ...options.reasoning ?? _MODEL?.reasoning ? {
             thinking: options.thinking || { type: 'enabled', budget_tokens: 1024 },
         } : {}, ..._MODEL?.tools ? {  // https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking
             tools: options.tools ?? toolsClaude.map(x => x.def),
             tool_choice: { type: 'auto' },
+            betas: ['token-efficient-tools-2025-02-19'],
+            // @todo: https://docs.anthropic.com/en/docs/build-with-claude/tool-use/token-efficient-tool-use
         } : {},
     });
     for await (const chunk of resp) {
@@ -1048,6 +1074,8 @@ const promptClaude = async (content, options = {}) => {
             tool_use[tool_use.length - 1].input += event.partial_json;
         }
         deltaText = deltaThink + deltaText;
+        options.result && deltaText
+            && (responded = responded || (deltaText = `\n\n${deltaText}`));
         result += deltaText;
         deltaText && await streamResp({
             text: options.delta ? deltaText : result,
@@ -1143,7 +1171,6 @@ const promptGemini = async (content, options = {}) => {
         ) : {},
     });
     // https://github.com/google/generative-ai-js/blob/main/samples/node/advanced-chat.js
-    // @todo: check this issue similar to Vertex AI:
     // Google's bug: history is not allowed while using inline_data?
     const chat = client.startChat({ history, ...generationConfig(options) });
     const resp = await chat.sendMessageStream(prompt);
@@ -1508,9 +1535,8 @@ export {
     ATTACHMENT_TOKEN_COST, CLOUD_37_SONNET, CODE_INTERPRETER, DEEPSEEK_R1,
     DEEPSEEK_R1_32B, DEEPSEEK_R1_70B, DEFAULT_MODELS,
     EMBEDDING_001,
-    FUNCTION, GEMINI_20_FLASH, GEMINI_20_FLASH_THINKING, GPT_4O, GPT_4O_MINI, GPT_O1, GPT_O3_MINI, INSTRUCTIONS, MODELS,
-    OPENAI_VOICE,
-    RETRIEVAL,
+    FUNCTION, GEMINI_20_FLASH, GEMINI_20_FLASH_THINKING, GPT_45, GPT_4O, GPT_4O_MINI, GPT_O1, GPT_O3_MINI, INSTRUCTIONS, MODELS,
+    OPENAI_VOICE, RETRIEVAL,
     TEXT_EMBEDDING_3_SMALL, _NEED, analyzeSessions,
     buildGptTrainingCase,
     buildGptTrainingCases,

package/lib/manifest.mjs

@@ -1,7 +1,7 @@
 const manifest = {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.42",
+    "version": "1998.2.44",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",

package/package.json

@@ -1,7 +1,7 @@
 {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.42",
+    "version": "1998.2.44",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",