mirror of
https://github.com/HeyPuter/puter.git
synced 2026-05-04 08:30:39 +00:00
fix: open router cache token metering (#1988)
This commit is contained in:
Daniel Salazar
Generated
+27
-2
@@ -904,6 +904,7 @@
|
||||
"integrity": "sha512-2BCOP7TN8M+gVDj7/ht3hsaO/B/n5oDbiAyyvnRlNOs+u1o+JWNYTQrmpuNp1/Wq2gcFrI01JAW+paEKDMx/CA==",
|
||||
"dev": true,
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@babel/code-frame": "^7.27.1",
|
||||
"@babel/generator": "^7.28.3",
|
||||
@@ -3047,6 +3048,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/custom/-/custom-0.22.12.tgz",
|
||||
"integrity": "sha512-xcmww1O/JFP2MrlGUMd3Q78S3Qu6W3mYTXYuIqFq33EorgYHV/HqymHfXy9GjiCJ7OI+7lWx6nYFOzU7M4rd1Q==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/core": "^0.22.12"
|
||||
}
|
||||
@@ -3083,6 +3085,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/plugin-blit/-/plugin-blit-0.22.12.tgz",
|
||||
"integrity": "sha512-xslz2ZoFZOPLY8EZ4dC29m168BtDx95D6K80TzgUi8gqT7LY6CsajWO0FAxDwHz6h0eomHMfyGX0stspBrTKnQ==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/utils": "^0.22.12"
|
||||
},
|
||||
@@ -3095,6 +3098,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/plugin-blur/-/plugin-blur-0.22.12.tgz",
|
||||
"integrity": "sha512-S0vJADTuh1Q9F+cXAwFPlrKWzDj2F9t/9JAbUvaaDuivpyWuImEKXVz5PUZw2NbpuSHjwssbTpOZ8F13iJX4uw==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/utils": "^0.22.12"
|
||||
},
|
||||
@@ -3119,6 +3123,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/plugin-color/-/plugin-color-0.22.12.tgz",
|
||||
"integrity": "sha512-xImhTE5BpS8xa+mAN6j4sMRWaUgUDLoaGHhJhpC+r7SKKErYDR0WQV4yCE4gP+N0gozD0F3Ka1LUSaMXrn7ZIA==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/utils": "^0.22.12",
|
||||
"tinycolor2": "^1.6.0"
|
||||
@@ -3162,6 +3167,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/plugin-crop/-/plugin-crop-0.22.12.tgz",
|
||||
"integrity": "sha512-FNuUN0OVzRCozx8XSgP9MyLGMxNHHJMFt+LJuFjn1mu3k0VQxrzqbN06yIl46TVejhyAhcq5gLzqmSCHvlcBVw==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/utils": "^0.22.12"
|
||||
},
|
||||
@@ -3285,6 +3291,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/plugin-resize/-/plugin-resize-0.22.12.tgz",
|
||||
"integrity": "sha512-3NyTPlPbTnGKDIbaBgQ3HbE6wXbAlFfxHVERmrbqAi8R3r6fQPxpCauA8UVDnieg5eo04D0T8nnnNIX//i/sXg==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/utils": "^0.22.12"
|
||||
},
|
||||
@@ -3297,6 +3304,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/plugin-rotate/-/plugin-rotate-0.22.12.tgz",
|
||||
"integrity": "sha512-9YNEt7BPAFfTls2FGfKBVgwwLUuKqy+E8bDGGEsOqHtbuhbshVGxN2WMZaD4gh5IDWvR+emmmPPWGgaYNYt1gA==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/utils": "^0.22.12"
|
||||
},
|
||||
@@ -3312,6 +3320,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@jimp/plugin-scale/-/plugin-scale-0.22.12.tgz",
|
||||
"integrity": "sha512-dghs92qM6MhHj0HrV2qAwKPMklQtjNpoYgAB94ysYpsXslhRTiPisueSIELRwZGEr0J0VUxpUY7HgJwlSIgGZw==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@jimp/utils": "^0.22.12"
|
||||
},
|
||||
@@ -3619,6 +3628,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@opentelemetry/api/-/api-1.4.1.tgz",
|
||||
"integrity": "sha512-O2yRJce1GOc6PAy3QxFM4NzFiWzvScDC1/5ihYBL6BUEVdq0XMWN01sppE+H6bBXbaFYipjwFLEWLg5PaSOThA==",
|
||||
"license": "Apache-2.0",
|
||||
"peer": true,
|
||||
"engines": {
|
||||
"node": ">=8.0.0"
|
||||
}
|
||||
@@ -3628,6 +3638,7 @@
|
||||
"resolved": "https://registry.npmjs.org/@opentelemetry/api-logs/-/api-logs-0.49.1.tgz",
|
||||
"integrity": "sha512-kaNl/T7WzyMUQHQlVq7q0oV4Kev6+0xFwqzofryC66jgGMacd0QH5TwfpbUwSTby+SdAdprAe5UKMvBw4tKS5Q==",
|
||||
"license": "Apache-2.0",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@opentelemetry/api": "^1.0.0"
|
||||
},
|
||||
@@ -7262,6 +7273,7 @@
|
||||
"integrity": "sha512-6JSSaBZmsKvEkbRUkf7Zj7dru/8ZCrJxAqArcLaVMee5907JdtEbKGsZ7zNiIm/UAkpGUkaSMZEXShnN2D1HZA==",
|
||||
"dev": true,
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@typescript-eslint/scope-manager": "8.46.1",
|
||||
"@typescript-eslint/types": "8.46.1",
|
||||
@@ -7804,7 +7816,8 @@
|
||||
"version": "5.5.0",
|
||||
"resolved": "https://registry.npmjs.org/@xterm/xterm/-/xterm-5.5.0.tgz",
|
||||
"integrity": "sha512-hqJHYaQb5OptNunnyAnkHyM8aCjZ1MEIDTQu1iIbbTD/xops91NB5yq1ZK/dC2JDbVWtF23zUtl9JE2NqwT87A==",
|
||||
"license": "MIT"
|
||||
"license": "MIT",
|
||||
"peer": true
|
||||
},
|
||||
"node_modules/@xtuc/ieee754": {
|
||||
"version": "1.2.0",
|
||||
@@ -7854,6 +7867,7 @@
|
||||
"resolved": "https://registry.npmjs.org/acorn/-/acorn-8.15.0.tgz",
|
||||
"integrity": "sha512-NZyJarBfL7nWwIq+FDL6Zp/yHEhePMNnnJ0y3qfieCrmNvYct8uvtiV41UvlSe6apAfk0fY1FbWx+NwfmpvtTg==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"bin": {
|
||||
"acorn": "bin/acorn"
|
||||
},
|
||||
@@ -8543,6 +8557,7 @@
|
||||
}
|
||||
],
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"baseline-browser-mapping": "^2.8.9",
|
||||
"caniuse-lite": "^1.0.30001746",
|
||||
@@ -8797,6 +8812,7 @@
|
||||
"resolved": "https://registry.npmjs.org/chai/-/chai-5.3.3.tgz",
|
||||
"integrity": "sha512-4zNhdJD/iOjSH0A05ea+Ke6MU5mmpQcbQsSOkgdaUMJ9zTlDTD/GYlwohmIE2u0gaxHYiVHEn1Fw9mZ/ktJWgw==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"assertion-error": "^2.0.1",
|
||||
"check-error": "^2.1.1",
|
||||
@@ -10533,6 +10549,7 @@
|
||||
"resolved": "https://registry.npmjs.org/eslint/-/eslint-9.37.0.tgz",
|
||||
"integrity": "sha512-XyLmROnACWqSxiGYArdef1fItQd47weqB7iwtfr9JHwRrqIXZdcFMvvEcL9xHCmL0SNsOvF0c42lWyM1U5dgig==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@eslint-community/eslint-utils": "^4.8.0",
|
||||
"@eslint-community/regexpp": "^4.12.1",
|
||||
@@ -16305,6 +16322,7 @@
|
||||
"resolved": "https://registry.npmjs.org/ajv/-/ajv-8.17.1.tgz",
|
||||
"integrity": "sha512-B/gBuNg5SiMTrPkC+A2+cW0RszwxYmn6VYxB/inlBStS5nx6xHIt/ehKRhIMhqusl7a8LjQoZnjCs5vhwxOQ1g==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"fast-deep-equal": "^3.1.3",
|
||||
"fast-uri": "^3.0.1",
|
||||
@@ -17985,6 +18003,7 @@
|
||||
"resolved": "https://registry.npmjs.org/typescript/-/typescript-5.9.3.tgz",
|
||||
"integrity": "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==",
|
||||
"license": "Apache-2.0",
|
||||
"peer": true,
|
||||
"bin": {
|
||||
"tsc": "bin/tsc",
|
||||
"tsserver": "bin/tsserver"
|
||||
@@ -18211,6 +18230,7 @@
|
||||
"integrity": "sha512-4nVGliEpxmhCL8DslSAUdxlB6+SMrhB0a1v5ijlh1xB1nEPuy1mxaHxysVucLHuWryAxLWg6a5ei+U4TLn/rFg==",
|
||||
"dev": true,
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"esbuild": "^0.25.0",
|
||||
"fdir": "^6.5.0",
|
||||
@@ -18476,6 +18496,7 @@
|
||||
"resolved": "https://registry.npmjs.org/webpack/-/webpack-5.102.1.tgz",
|
||||
"integrity": "sha512-7h/weGm9d/ywQ6qzJ+Xy+r9n/3qgp/thalBbpOi5i223dPXKi04IBtqPN9nTd+jBc7QKfvDbaBnFipYp4sJAUQ==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@types/eslint-scope": "^3.7.7",
|
||||
"@types/estree": "^1.0.8",
|
||||
@@ -18525,6 +18546,7 @@
|
||||
"integrity": "sha512-pIDJHIEI9LR0yxHXQ+Qh95k2EvXpWzZ5l+d+jIo+RdSm9MiHfzazIxwwni/p7+x4eJZuvG1AJwgC4TNQ7NRgsg==",
|
||||
"dev": true,
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@discoveryjs/json-ext": "^0.5.0",
|
||||
"@webpack-cli/configtest": "^2.1.1",
|
||||
@@ -18756,6 +18778,7 @@
|
||||
"resolved": "https://registry.npmjs.org/winston/-/winston-3.18.3.tgz",
|
||||
"integrity": "sha512-NoBZauFNNWENgsnC9YpgyYwOVrl2m58PpQ8lNHjV3kosGs7KJ7Npk9pCUE+WJlawVSe8mykWDKWFSVfs3QO9ww==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"dependencies": {
|
||||
"@colors/colors": "^1.6.0",
|
||||
"@dabh/diagnostics": "^2.0.8",
|
||||
@@ -18960,6 +18983,7 @@
|
||||
"resolved": "https://registry.npmjs.org/ws/-/ws-8.18.3.tgz",
|
||||
"integrity": "sha512-PEIGCY5tSlUt50cqyMXfCzX+oOPqN0vuGqWzbcJ2xvnkzkq46oOpz7dQaTDBdfICb4N14+GARUDw2XV2N4tvzg==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"engines": {
|
||||
"node": ">=10.0.0"
|
||||
},
|
||||
@@ -19293,6 +19317,7 @@
|
||||
"resolved": "https://registry.npmjs.org/zod/-/zod-3.25.76.tgz",
|
||||
"integrity": "sha512-gzUt/qt81nXsFGKIFcC3YnfEAx5NkunCfnDlvuBSSFS02bcXu4Lmea0AFIUwbLWxWPx3d9p8S5QoaujKcNQxcQ==",
|
||||
"license": "MIT",
|
||||
"peer": true,
|
||||
"funding": {
|
||||
"url": "https://github.com/sponsors/colinhacks"
|
||||
}
|
||||
@@ -20068,7 +20093,7 @@
|
||||
},
|
||||
"src/puter-js": {
|
||||
"name": "@heyputer/puter.js",
|
||||
"version": "2.1.2",
|
||||
"version": "2.1.6",
|
||||
"license": "Apache-2.0",
|
||||
"dependencies": {
|
||||
"@heyputer/kv.js": "^0.2.1",
|
||||
|
||||
@@ -25,6 +25,7 @@ const Messages = require('./lib/Messages');
|
||||
const FSNodeParam = require('../../api/filesystem/FSNodeParam');
|
||||
const { LLRead } = require('../../filesystem/ll_operations/ll_read');
|
||||
const { Context } = require('../../util/context');
|
||||
const mime = require('mime-types');
|
||||
|
||||
/**
|
||||
* ClaudeService class extends BaseService to provide integration with Anthropic's Claude AI models.
|
||||
@@ -34,10 +35,6 @@ const { Context } = require('../../util/context');
|
||||
* @extends BaseService
|
||||
*/
|
||||
class ClaudeService extends BaseService {
|
||||
static MODULES = {
|
||||
Anthropic: require('@anthropic-ai/sdk'),
|
||||
};
|
||||
|
||||
/**
|
||||
* @type {import('@anthropic-ai/sdk').Anthropic}
|
||||
*/
|
||||
@@ -53,7 +50,7 @@ class ClaudeService extends BaseService {
|
||||
/** @type {import('../../services/MeteringService/MeteringService').MeteringService} */
|
||||
#meteringService;
|
||||
|
||||
async _init() {
|
||||
async _init () {
|
||||
this.anthropic = new Anthropic({
|
||||
apiKey: this.config.apiKey,
|
||||
// 10 minutes is the default; we need to override the timeout to
|
||||
@@ -75,7 +72,7 @@ class ClaudeService extends BaseService {
|
||||
* Returns the default model identifier for Claude API interactions
|
||||
* @returns {string} The default model ID 'claude-3-5-sonnet-latest'
|
||||
*/
|
||||
get_default_model() {
|
||||
get_default_model () {
|
||||
return 'claude-3-5-sonnet-latest';
|
||||
}
|
||||
|
||||
@@ -87,7 +84,7 @@ class ClaudeService extends BaseService {
|
||||
*
|
||||
* @returns Promise<Array<Object>> Array of model details
|
||||
*/
|
||||
async models() {
|
||||
async models () {
|
||||
return this.models_();
|
||||
},
|
||||
|
||||
@@ -97,7 +94,7 @@ class ClaudeService extends BaseService {
|
||||
* @description Retrieves all available model IDs and their aliases,
|
||||
* flattening them into a single array of strings that can be used for model selection
|
||||
*/
|
||||
async list() {
|
||||
async list () {
|
||||
const models = this.models_();
|
||||
const model_names = [];
|
||||
for ( const model of models ) {
|
||||
@@ -118,10 +115,10 @@ class ClaudeService extends BaseService {
|
||||
* @returns {Object} Returns either a TypedValue with streaming response or a completion object
|
||||
* @this {ClaudeService}
|
||||
*/
|
||||
async complete({ messages, stream, model, tools, max_tokens, temperature }) {
|
||||
async complete ({ messages, stream, model, tools, max_tokens, temperature }) {
|
||||
tools = FunctionCalling.make_claude_tools(tools);
|
||||
// console.log("here are the messages: ", messages)
|
||||
|
||||
|
||||
let system_prompts;
|
||||
// unsure why system_prompts is an array but it always seems to only have exactly one element,
|
||||
// and the real array of system_prompts seems to be the [0].content -- NS
|
||||
@@ -140,12 +137,12 @@ class ClaudeService extends BaseService {
|
||||
}
|
||||
|
||||
messages = messages.map(message => {
|
||||
if (message.cache_control) {
|
||||
if ( message.cache_control ) {
|
||||
message.content[0].cache_control = message.cache_control;
|
||||
}
|
||||
delete message.cache_control;
|
||||
return message
|
||||
})
|
||||
return message;
|
||||
});
|
||||
|
||||
const sdk_params = {
|
||||
model: model ?? this.get_default_model(),
|
||||
@@ -153,19 +150,19 @@ class ClaudeService extends BaseService {
|
||||
((
|
||||
model === 'claude-3-5-sonnet-20241022'
|
||||
|| model === 'claude-3-5-sonnet-20240620'
|
||||
) ? 8192 : this.models_().filter(e=>e.name === model)[0].max_tokens || 4096), //required
|
||||
) ? 8192 : this.models_().filter(e => e.name === model)[0].max_tokens || 4096), //required
|
||||
temperature: temperature || 0, // required
|
||||
...( (system_prompts && system_prompts[0]?.content) ? {
|
||||
system: system_prompts[0]?.content
|
||||
system: system_prompts[0]?.content,
|
||||
} : {}),
|
||||
tool_choice: {
|
||||
type: "auto",
|
||||
disable_parallel_tool_use: true
|
||||
type: 'auto',
|
||||
disable_parallel_tool_use: true,
|
||||
},
|
||||
messages,
|
||||
...(tools ? { tools } : {}),
|
||||
};
|
||||
console.log(sdk_params.max_tokens)
|
||||
console.log(sdk_params.max_tokens);
|
||||
|
||||
// console.log('\x1B[26;1m ===== SDK PARAMETERS', require('util').inspect(sdk_params, undefined, Infinity));
|
||||
|
||||
@@ -201,8 +198,6 @@ class ClaudeService extends BaseService {
|
||||
fsNode: task.node,
|
||||
});
|
||||
|
||||
const require = this.require;
|
||||
const mime = require('mime-types');
|
||||
const mimeType = mime.contentType(await task.node.get('name'));
|
||||
|
||||
beta_mode = true;
|
||||
@@ -228,12 +223,6 @@ class ClaudeService extends BaseService {
|
||||
return 'container_upload';
|
||||
})();
|
||||
|
||||
// {
|
||||
// 'application/pdf': 'document',
|
||||
// 'text/plain': 'document',
|
||||
// 'image/': 'image'
|
||||
// }[mimeType];
|
||||
|
||||
delete task.contentPart.puter_path,
|
||||
task.contentPart.type = contentBlockTypeForFileBasedOnMime;
|
||||
task.contentPart.source = {
|
||||
@@ -251,7 +240,7 @@ class ClaudeService extends BaseService {
|
||||
try {
|
||||
await this.anthropic.beta.files.delete(task.file_id,
|
||||
{ betas: ['files-api-2025-04-14'] });
|
||||
} catch (e) {
|
||||
} catch (e) {
|
||||
this.errors.report('claude:file-delete-task', {
|
||||
source: e,
|
||||
trace: true,
|
||||
@@ -328,26 +317,7 @@ class ClaudeService extends BaseService {
|
||||
}
|
||||
chatStream.end();
|
||||
|
||||
this.billForUsage(actor, model || this.get_default_model(), usageSum);
|
||||
|
||||
// Log token usage statistics
|
||||
const totalTokens = usageSum.input_tokens + usageSum.output_tokens;
|
||||
const cachedTokens = usageSum.ephemeral_5m_input_tokens + usageSum.ephemeral_1h_input_tokens;
|
||||
const cacheHits = usageSum.cache_read_input_tokens;
|
||||
const uncachedTokens = usageSum.input_tokens - cacheHits - cachedTokens;
|
||||
|
||||
// console.log(`
|
||||
// ╔══════════════════════════════════════════════════════════════╗
|
||||
// ║ 🎯 Token Usage Statistics 🎯 ║
|
||||
// ╠══════════════════════════════════════════════════════════════╣
|
||||
// ║ 📊 Total Tokens Used: ${String(totalTokens).padStart(10)} 📊 ║
|
||||
// ║ 💾 Cached Tokens: ${String(cachedTokens).padStart(10)} 💾 ║
|
||||
// ║ ✅ Cache Hits: ${String(cacheHits).padStart(10)} ✅ ║
|
||||
// ║ 🔄 Uncached Tokens: ${String(uncachedTokens).padStart(10)} 🔄 ║
|
||||
// ║ 📥 Input Tokens: ${String(usageSum.input_tokens).padStart(10)} 📥 ║
|
||||
// ║ 📤 Output Tokens: ${String(usageSum.output_tokens).padStart(10)} 📤 ║
|
||||
// ╚══════════════════════════════════════════════════════════════╝
|
||||
// `);
|
||||
this.#meteringService.utilRecordUsageObject(usageSum, actor, `claude:${this.models_().find(m => [m.id, ...(m.aliases || [])].includes(model || this.get_default_model())).id}`);
|
||||
};
|
||||
|
||||
return {
|
||||
@@ -361,26 +331,7 @@ class ClaudeService extends BaseService {
|
||||
await cleanup_files();
|
||||
|
||||
const usage = this.usageFormatterUtil(msg.usage);
|
||||
this.billForUsage(actor, model || this.get_default_model(), usage);
|
||||
|
||||
// Log token usage statistics
|
||||
const totalTokens = usage.input_tokens + usage.output_tokens;
|
||||
const cachedTokens = usage.ephemeral_5m_input_tokens + usage.ephemeral_1h_input_tokens;
|
||||
const cacheHits = usage.cache_read_input_tokens;
|
||||
const uncachedTokens = usage.input_tokens - cacheHits - cachedTokens;
|
||||
|
||||
// console.log(`
|
||||
// ╔══════════════════════════════════════════════════════════════╗
|
||||
// ║ 🎯 Token Usage Statistics 🎯 ║
|
||||
// ╠══════════════════════════════════════════════════════════════╣
|
||||
// ║ 📊 Total Tokens Used: ${String(totalTokens).padStart(10)} 📊 ║
|
||||
// ║ 💾 Cached Tokens: ${String(cachedTokens).padStart(10)} 💾 ║
|
||||
// ║ ✅ Cache Hits: ${String(cacheHits).padStart(10)} ✅ ║
|
||||
// ║ 🔄 Uncached Tokens: ${String(uncachedTokens).padStart(10)} 🔄 ║
|
||||
// ║ 📥 Input Tokens: ${String(usage.input_tokens).padStart(10)} 📥 ║
|
||||
// ║ 📤 Output Tokens: ${String(usage.output_tokens).padStart(10)} 📤 ║
|
||||
// ╚══════════════════════════════════════════════════════════════╝
|
||||
// `);
|
||||
this.#meteringService.utilRecordUsageObject(usage, actor, `claude:${this.models_().find(m => [m.id, ...(m.aliases || [])].includes(model || this.get_default_model())).id}`);
|
||||
|
||||
// TODO DS: cleanup old usage tracking
|
||||
return {
|
||||
@@ -394,7 +345,7 @@ class ClaudeService extends BaseService {
|
||||
|
||||
// TODO DS: get this inside the class as a private method once the methods aren't exported directly
|
||||
/** @type {(usage: import("@anthropic-ai/sdk/resources/messages.js").Usage | import("@anthropic-ai/sdk/resources/beta/messages/messages.js").BetaUsage) => {}}) */
|
||||
usageFormatterUtil(usage) {
|
||||
usageFormatterUtil (usage) {
|
||||
return {
|
||||
input_tokens: usage?.input_tokens || 0,
|
||||
ephemeral_5m_input_tokens: usage?.cache_creation?.ephemeral_5m_input_tokens || usage.cache_creation_input_tokens || 0, // this is because they're api is a bit inconsistent
|
||||
@@ -404,11 +355,6 @@ class ClaudeService extends BaseService {
|
||||
};
|
||||
};
|
||||
|
||||
// TODO DS: get this inside the class as a private method once the methods aren't exported directly
|
||||
billForUsage(actor, model, usage) {
|
||||
this.#meteringService.utilRecordUsageObject(usage, actor, `claude:${this.models_().find(m => [m.id, ...(m.aliases || [])].includes(model)).id}`);
|
||||
};
|
||||
|
||||
/**
|
||||
* Retrieves available Claude AI models and their specifications
|
||||
* @returns Array of model objects containing:
|
||||
@@ -421,7 +367,7 @@ class ClaudeService extends BaseService {
|
||||
* - max_output: Maximum output tokens
|
||||
* - training_cutoff: Training data cutoff date
|
||||
*/
|
||||
models_() {
|
||||
models_ () {
|
||||
return [
|
||||
{
|
||||
id: 'claude-haiku-4-5-20251001',
|
||||
|
||||
@@ -18,11 +18,12 @@
|
||||
*/
|
||||
|
||||
// METADATA // {"ai-commented":{"service":"claude"}}
|
||||
const APIError = require('../../api/APIError');
|
||||
const BaseService = require('../../services/BaseService');
|
||||
const OpenAIUtil = require('./lib/OpenAIUtil');
|
||||
const { Context } = require('../../util/context');
|
||||
|
||||
const openai = require('openai');
|
||||
const uuidv4 = require('uuid').v4;
|
||||
const axios = require('axios');
|
||||
/**
|
||||
* OllamaService class - Provides integration with Ollama's API for chat completions
|
||||
* Extends BaseService to implement the puter-chat-completion interface.
|
||||
@@ -32,17 +33,14 @@ const { Context } = require('../../util/context');
|
||||
*/
|
||||
class OllamaService extends BaseService {
|
||||
static MODULES = {
|
||||
openai: require('openai'),
|
||||
kv: globalThis.kv,
|
||||
uuidv4: require('uuid').v4,
|
||||
axios: require('axios'),
|
||||
};
|
||||
|
||||
/**
|
||||
* Gets the system prompt used for AI interactions
|
||||
* @returns {string} The base system prompt that identifies the AI as running on Puter
|
||||
*/
|
||||
adapt_model(model) {
|
||||
adapt_model (model) {
|
||||
return model;
|
||||
}
|
||||
|
||||
@@ -51,16 +49,16 @@ class OllamaService extends BaseService {
|
||||
* @private
|
||||
* @returns {Promise<void>} Resolves when initialization is complete
|
||||
*/
|
||||
async _init() {
|
||||
async _init () {
|
||||
// Ollama typically runs on HTTP, not HTTPS
|
||||
this.api_base_url = this.config?.api_base_url || 'http://localhost:11434';
|
||||
|
||||
// OpenAI SDK is used to interact with the Ollama API
|
||||
this.openai = new this.modules.openai.OpenAI({
|
||||
apiKey: "ollama", // Ollama doesn't use an API key, it uses the "ollama" string
|
||||
baseURL: this.api_base_url + '/v1',
|
||||
this.openai = new openai.OpenAI({
|
||||
apiKey: 'ollama', // Ollama doesn't use an API key, it uses the "ollama" string
|
||||
baseURL: `${this.api_base_url }/v1`,
|
||||
});
|
||||
this.kvkey = this.modules.uuidv4();
|
||||
this.kvkey = uuidv4();
|
||||
|
||||
const svc_aiChat = this.services.get('ai-chat');
|
||||
svc_aiChat.register_provider({
|
||||
@@ -74,7 +72,7 @@ class OllamaService extends BaseService {
|
||||
* Returns the default model identifier for the Ollama service
|
||||
* @returns {string} The default model ID 'gpt-oss:20b'
|
||||
*/
|
||||
get_default_model() {
|
||||
get_default_model () {
|
||||
return 'gpt-oss:20b';
|
||||
}
|
||||
|
||||
@@ -86,7 +84,7 @@ class OllamaService extends BaseService {
|
||||
*
|
||||
* @returns Promise<Array<Object>> Array of model details
|
||||
*/
|
||||
async models() {
|
||||
async models () {
|
||||
return await this.models_();
|
||||
},
|
||||
/**
|
||||
@@ -95,7 +93,7 @@ class OllamaService extends BaseService {
|
||||
* @description Retrieves all available model IDs and their aliases,
|
||||
* flattening them into a single array of strings that can be used for model selection
|
||||
*/
|
||||
async list() {
|
||||
async list () {
|
||||
const models = await this.models_();
|
||||
const model_names = [];
|
||||
for ( const model of models ) {
|
||||
@@ -108,7 +106,7 @@ class OllamaService extends BaseService {
|
||||
* AI Chat completion method.
|
||||
* See AIChatService for more details.
|
||||
*/
|
||||
async complete({ messages, stream, model, tools, max_tokens, temperature }) {
|
||||
async complete ({ messages, stream, model, tools, max_tokens, temperature }) {
|
||||
model = this.adapt_model(model);
|
||||
|
||||
if ( model.startsWith('ollama:') ) {
|
||||
@@ -128,19 +126,23 @@ class OllamaService extends BaseService {
|
||||
...(stream ? {
|
||||
stream_options: { include_usage: true },
|
||||
} : {}),
|
||||
}
|
||||
};
|
||||
|
||||
const completion = await this.openai.chat.completions.create(sdk_params);
|
||||
|
||||
const modelDetails = (await this.models_()).find(m => m.id === 'ollama:' + model);
|
||||
const modelDetails = (await this.models_()).find(m => m.id === `ollama:${model}`);
|
||||
return OpenAIUtil.handle_completion_output({
|
||||
usage_calculator: ({ usage }) => {
|
||||
// custom open router logic because its free
|
||||
|
||||
const trackedUsage = {
|
||||
prompt: 0,
|
||||
completion: 0,
|
||||
input_cache_read: 0,
|
||||
prompt: (usage.prompt_tokens ?? 1 ) - (usage.prompt_tokens_details?.cached_tokens ?? 0),
|
||||
completion: usage.completion_tokens ?? 1,
|
||||
input_cache_read: usage.prompt_tokens_details?.cached_tokens ?? 0,
|
||||
};
|
||||
const costOverwrites = Object.fromEntries(Object.keys(trackedUsage).map((k) => {
|
||||
return [k, 0]; // override to 0 since local is free
|
||||
}));
|
||||
this.meteringService.utilRecordUsageObject(trackedUsage, actor, modelDetails.id, { costOverwrites });
|
||||
const legacyCostCalculator = OpenAIUtil.create_usage_calculator({
|
||||
model_details: modelDetails,
|
||||
});
|
||||
@@ -162,31 +164,30 @@ class OllamaService extends BaseService {
|
||||
* - cost: Pricing information object with currency and rates
|
||||
* @private
|
||||
*/
|
||||
async models_(rawPriceKeys = false) {
|
||||
const axios = this.require('axios');
|
||||
async models_ (_rawPriceKeys = false) {
|
||||
|
||||
let models = this.modules.kv.get(`${this.kvkey}:models`);
|
||||
if ( !models ) {
|
||||
if ( ! models ) {
|
||||
try {
|
||||
const resp = await axios.request({
|
||||
method: 'GET',
|
||||
url: this.api_base_url + '/api/tags',
|
||||
url: `${this.api_base_url}/api/tags`,
|
||||
});
|
||||
models = resp.data.models || [];
|
||||
if ( models.length > 0 ) {
|
||||
this.modules.kv.set(`${this.kvkey}:models`, models);
|
||||
}
|
||||
} catch (error) {
|
||||
} catch ( error ) {
|
||||
this.log.error('Failed to fetch models from Ollama:', error.message);
|
||||
// Return empty array if Ollama is not available
|
||||
return [];
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
if ( !models || models.length === 0 ) {
|
||||
return [];
|
||||
}
|
||||
|
||||
|
||||
const coerced_models = [];
|
||||
for ( const model of models ) {
|
||||
// Ollama API returns models with 'name' property, not 'model'
|
||||
@@ -196,8 +197,8 @@ class OllamaService extends BaseService {
|
||||
output: 0,
|
||||
};
|
||||
coerced_models.push({
|
||||
id: 'ollama:' + modelName,
|
||||
name: modelName + ' (Ollama)',
|
||||
id: `ollama:${ modelName}`,
|
||||
name: `${modelName} (Ollama)`,
|
||||
max_tokens: model.size || model.max_context || 8192,
|
||||
cost: {
|
||||
currency: 'usd-cents',
|
||||
@@ -206,7 +207,7 @@ class OllamaService extends BaseService {
|
||||
},
|
||||
});
|
||||
}
|
||||
console.log("coerced_models", coerced_models);
|
||||
console.log('coerced_models', coerced_models);
|
||||
return coerced_models;
|
||||
}
|
||||
}
|
||||
|
||||
+12
-12
@@ -51,7 +51,7 @@ export class OpenAICompletionService {
|
||||
/** @type {import('../../../services/MeteringService/MeteringService.js').MeteringService} */
|
||||
#meteringService;
|
||||
|
||||
constructor({ serviceName, config, globalConfig, aiChatService, meteringService, models = OPEN_AI_MODELS, defaultModel = 'gpt-5-nano' }) {
|
||||
constructor ({ serviceName, config, globalConfig, aiChatService, meteringService, models = OPEN_AI_MODELS, defaultModel = 'gpt-5-nano' }) {
|
||||
this.#models = models;
|
||||
this.#defaultModel = defaultModel;
|
||||
this.#meteringService = meteringService;
|
||||
@@ -60,7 +60,7 @@ export class OpenAICompletionService {
|
||||
globalConfig?.services?.openai?.apiKey;
|
||||
|
||||
// Fallback to the old format for backward compatibility
|
||||
if ( !apiKey ) {
|
||||
if ( ! apiKey ) {
|
||||
apiKey =
|
||||
config?.openai?.secret_key ??
|
||||
globalConfig?.openai?.secret_key;
|
||||
@@ -70,7 +70,7 @@ export class OpenAICompletionService {
|
||||
'Please use `services.openai.apiKey` instead.');
|
||||
}
|
||||
|
||||
if ( !apiKey ) {
|
||||
if ( ! apiKey ) {
|
||||
throw new Error('OpenAI API key is missing in configuration.');
|
||||
}
|
||||
|
||||
@@ -89,11 +89,11 @@ export class OpenAICompletionService {
|
||||
* Each model object includes an ID and cost details (currency, tokens, input/output rates).
|
||||
* @returns {{id: string, cost: {currency: string, tokens: number, input: number, output: number}}[]}
|
||||
*/
|
||||
models() {
|
||||
models () {
|
||||
return this.#models;
|
||||
}
|
||||
|
||||
list() {
|
||||
list () {
|
||||
const models = this.models();
|
||||
const model_names = [];
|
||||
for ( const model of models ) {
|
||||
@@ -105,11 +105,11 @@ export class OpenAICompletionService {
|
||||
return model_names;
|
||||
}
|
||||
|
||||
get_default_model(){
|
||||
get_default_model () {
|
||||
return this.#defaultModel;
|
||||
}
|
||||
|
||||
async complete({ messages, stream, model, tools, max_tokens, temperature, reasoning, text, reasoning_effort, verbosity }) {
|
||||
async complete ({ messages, stream, model, tools, max_tokens, temperature, reasoning, text, reasoning_effort, verbosity }) {
|
||||
return await this.#complete(messages, {
|
||||
model: model,
|
||||
tools,
|
||||
@@ -131,20 +131,20 @@ export class OpenAICompletionService {
|
||||
* @property {boolean} flagged - Whether the content was flagged as inappropriate
|
||||
* @property {Object} results - Raw moderation results from OpenAI API
|
||||
*/
|
||||
async checkModeration(text) {
|
||||
async checkModeration (text) {
|
||||
// create moderation
|
||||
const results = await this.#openAi.moderations.create({
|
||||
model: "omni-moderation-latest",
|
||||
model: 'omni-moderation-latest',
|
||||
input: text,
|
||||
});
|
||||
|
||||
let flagged = false;
|
||||
|
||||
for ( const result of results?.results ?? [] ) {
|
||||
|
||||
|
||||
// OpenAI does a crazy amount of false positives. We filter by their 80% interval
|
||||
const veryFlaggedEntries = Object.entries(result.category_scores).filter(e => e[1] > 0.8);
|
||||
if (veryFlaggedEntries.length > 0 ) {
|
||||
if ( veryFlaggedEntries.length > 0 ) {
|
||||
flagged = true;
|
||||
break;
|
||||
}
|
||||
@@ -166,7 +166,7 @@ export class OpenAICompletionService {
|
||||
* @returns {Promise<Object>} The completion response containing message and usage info
|
||||
* @throws {Error} If messages are invalid or content is flagged by moderation
|
||||
*/
|
||||
async #complete(messages, {
|
||||
async #complete (messages, {
|
||||
stream, moderation, model, tools,
|
||||
temperature, max_tokens,
|
||||
reasoning, text, reasoning_effort, verbosity,
|
||||
|
||||
@@ -26,7 +26,7 @@ export class OpenAICompletionServiceWrapper extends BaseService {
|
||||
/** @type {OpenAICompletionService} */
|
||||
openAICompletionService;
|
||||
|
||||
_init(){
|
||||
_init () {
|
||||
this.openAICompletionService = new OpenAICompletionService({
|
||||
serviceName: this.service_name,
|
||||
config: this.config,
|
||||
@@ -36,12 +36,12 @@ export class OpenAICompletionServiceWrapper extends BaseService {
|
||||
});
|
||||
}
|
||||
|
||||
async check_moderation(text) {
|
||||
async check_moderation (text) {
|
||||
return await this.openAICompletionService.checkModeration(text);
|
||||
}
|
||||
|
||||
get_default_model() {
|
||||
return this.openAICompletionService.get_default_model();
|
||||
get_default_model () {
|
||||
return this.openAICompletionService.get_default_model();
|
||||
}
|
||||
|
||||
static IMPLEMENTS = {
|
||||
@@ -49,7 +49,7 @@ export class OpenAICompletionServiceWrapper extends BaseService {
|
||||
.filter(n => n !== 'constructor')
|
||||
.reduce((acc, fn) => ({
|
||||
...acc,
|
||||
[fn]: async function(...a) {
|
||||
[fn]: async function (...a) {
|
||||
return await this.openAICompletionService[fn](...a);
|
||||
},
|
||||
}), {}),
|
||||
|
||||
@@ -11,7 +11,7 @@ export const OPEN_AI_MODELS = [
|
||||
},
|
||||
max_tokens: 128000,
|
||||
},
|
||||
{
|
||||
{
|
||||
id: 'gpt-5.1-codex',
|
||||
cost: {
|
||||
currency: 'usd-cents',
|
||||
|
||||
@@ -22,7 +22,9 @@ const APIError = require('../../api/APIError');
|
||||
const BaseService = require('../../services/BaseService');
|
||||
const OpenAIUtil = require('./lib/OpenAIUtil');
|
||||
const { Context } = require('../../util/context');
|
||||
|
||||
const openai = require('openai');
|
||||
const uuidv4 = require('uuid').v4;
|
||||
const axios = require('axios');
|
||||
/**
|
||||
* XAIService class - Provides integration with X.AI's API for chat completions
|
||||
* Extends BaseService to implement the puter-chat-completion interface.
|
||||
@@ -32,17 +34,29 @@ const { Context } = require('../../util/context');
|
||||
*/
|
||||
class OpenRouterService extends BaseService {
|
||||
static MODULES = {
|
||||
openai: require('openai'),
|
||||
kv: globalThis.kv,
|
||||
uuidv4: require('uuid').v4,
|
||||
axios: require('axios'),
|
||||
};
|
||||
|
||||
// TODO DS: extract this into driver wrapper like openAiService
|
||||
static IMPLEMENTS = {
|
||||
['puter-chat-completion']: {
|
||||
async models () {
|
||||
return await this.models();
|
||||
},
|
||||
async list () {
|
||||
return await this.list();
|
||||
},
|
||||
async complete (...params) {
|
||||
return await this.complete(...params);
|
||||
},
|
||||
},
|
||||
};
|
||||
|
||||
/**
|
||||
* Gets the system prompt used for AI interactions
|
||||
* @returns {string} The base system prompt that identifies the AI as running on Puter
|
||||
*/
|
||||
adapt_model(model) {
|
||||
adapt_model (model) {
|
||||
return model;
|
||||
}
|
||||
|
||||
@@ -54,13 +68,13 @@ class OpenRouterService extends BaseService {
|
||||
* @private
|
||||
* @returns {Promise<void>} Resolves when initialization is complete
|
||||
*/
|
||||
async _init() {
|
||||
async _init () {
|
||||
this.api_base_url = 'https://openrouter.ai/api/v1';
|
||||
this.openai = new this.modules.openai.OpenAI({
|
||||
this.openai = new openai.OpenAI({
|
||||
apiKey: this.config.apiKey,
|
||||
baseURL: this.api_base_url,
|
||||
});
|
||||
this.kvkey = this.modules.uuidv4();
|
||||
this.kvkey = uuidv4();
|
||||
|
||||
const svc_aiChat = this.services.get('ai-chat');
|
||||
svc_aiChat.register_provider({
|
||||
@@ -74,97 +88,81 @@ class OpenRouterService extends BaseService {
|
||||
* Returns the default model identifier for the XAI service
|
||||
* @returns {string} The default model ID 'grok-beta'
|
||||
*/
|
||||
get_default_model() {
|
||||
return 'grok-beta';
|
||||
get_default_model () {
|
||||
}
|
||||
|
||||
static IMPLEMENTS = {
|
||||
['puter-chat-completion']: {
|
||||
/**
|
||||
* Returns a list of available models and their details.
|
||||
* See AIChatService for more information.
|
||||
*
|
||||
* @returns Promise<Array<Object>> Array of model details
|
||||
*/
|
||||
async models() {
|
||||
return await this.models_();
|
||||
},
|
||||
/**
|
||||
/**
|
||||
* Returns a list of available model names including their aliases
|
||||
* @returns {Promise<string[]>} Array of model identifiers and their aliases
|
||||
* @description Retrieves all available model IDs and their aliases,
|
||||
* flattening them into a single array of strings that can be used for model selection
|
||||
*/
|
||||
async list() {
|
||||
const models = await this.models_();
|
||||
const model_names = [];
|
||||
for ( const model of models ) {
|
||||
model_names.push(model.id);
|
||||
}
|
||||
return model_names;
|
||||
},
|
||||
async list () {
|
||||
const models = await this.models();
|
||||
const model_names = [];
|
||||
for ( const model of models ) {
|
||||
model_names.push(model.id);
|
||||
}
|
||||
return model_names;
|
||||
}
|
||||
|
||||
/**
|
||||
/**
|
||||
* AI Chat completion method.
|
||||
* See AIChatService for more details.
|
||||
*/
|
||||
async complete({ messages, stream, model, tools, max_tokens, temperature }) {
|
||||
model = this.adapt_model(model);
|
||||
async complete ({ messages, stream, model, tools, max_tokens, temperature }) {
|
||||
model = this.adapt_model(model);
|
||||
|
||||
if ( model.startsWith('openrouter:') ) {
|
||||
model = model.slice('openrouter:'.length);
|
||||
}
|
||||
if ( model.startsWith('openrouter:') ) {
|
||||
model = model.slice('openrouter:'.length);
|
||||
}
|
||||
|
||||
if ( model === 'openrouter/auto' ) {
|
||||
throw APIError.create('field_invalid', null, {
|
||||
key: 'model',
|
||||
expected: 'allowed model',
|
||||
got: 'disallowed model',
|
||||
});
|
||||
}
|
||||
if ( model === 'openrouter/auto' ) {
|
||||
throw APIError.create('field_invalid', null, {
|
||||
key: 'model',
|
||||
expected: 'allowed model',
|
||||
got: 'disallowed model',
|
||||
});
|
||||
}
|
||||
|
||||
const actor = Context.get('actor');
|
||||
const actor = Context.get('actor');
|
||||
|
||||
messages = await OpenAIUtil.process_input_messages(messages);
|
||||
const sdk_params = {
|
||||
messages,
|
||||
model: model ?? this.get_default_model(),
|
||||
...(tools ? { tools } : {}),
|
||||
max_tokens,
|
||||
temperature: temperature, // default to 1.0
|
||||
stream,
|
||||
...(stream ? {
|
||||
stream_options: { include_usage: true },
|
||||
} : {}),
|
||||
}
|
||||
messages = await OpenAIUtil.process_input_messages(messages);
|
||||
|
||||
const completion = await this.openai.chat.completions.create(sdk_params);
|
||||
const completion = await this.openai.chat.completions.create({
|
||||
messages,
|
||||
model: model ?? this.get_default_model(),
|
||||
...(tools ? { tools } : {}),
|
||||
max_tokens,
|
||||
temperature: temperature, // default to 1.0
|
||||
stream,
|
||||
...(stream ? {
|
||||
stream_options: { include_usage: true },
|
||||
} : {}),
|
||||
});
|
||||
|
||||
const modelDetails = (await this.models_()).find(m => m.id === 'openrouter:' + model);
|
||||
const rawPriceModelDetails = (await this.models_(true)).find(m => m.id === 'openrouter:' + model);
|
||||
return OpenAIUtil.handle_completion_output({
|
||||
usage_calculator: ({ usage }) => {
|
||||
// custom open router logic because they're pricing are weird
|
||||
const trackedUsage = {
|
||||
prompt: usage.prompt_tokens ?? 0,
|
||||
completion: usage.completion_tokens ?? 0,
|
||||
input_cache_read: usage.prompt_tokens_details?.cached_tokens ?? 0,
|
||||
};
|
||||
const costOverwrites = Object.fromEntries(Object.keys(trackedUsage).map((k) => {
|
||||
return [k, rawPriceModelDetails.cost[k] * trackedUsage[k]];
|
||||
}));
|
||||
this.meteringService.utilRecordUsageObject(trackedUsage, actor, modelDetails.id, costOverwrites);
|
||||
const legacyCostCalculator = OpenAIUtil.create_usage_calculator({
|
||||
model_details: modelDetails,
|
||||
});
|
||||
return legacyCostCalculator({ usage });
|
||||
},
|
||||
stream,
|
||||
completion,
|
||||
const modelDetails = (await this.models()).find(m => m.id === `openrouter:${ model}`);
|
||||
const rawPriceModelDetails = (await this.models(true)).find(m => m.id === `openrouter:${ model}`);
|
||||
return OpenAIUtil.handle_completion_output({
|
||||
usage_calculator: ({ usage }) => {
|
||||
// custom open router logic because they're pricing are weird
|
||||
const trackedUsage = {
|
||||
prompt: (usage.prompt_tokens ?? 0 ) - (usage.prompt_tokens_details?.cached_tokens ?? 0),
|
||||
completion: usage.completion_tokens ?? 0,
|
||||
input_cache_read: usage.prompt_tokens_details?.cached_tokens ?? 0,
|
||||
};
|
||||
const costOverwrites = Object.fromEntries(Object.keys(trackedUsage).map((k) => {
|
||||
return [k, rawPriceModelDetails.cost[k] * trackedUsage[k]];
|
||||
}));
|
||||
this.meteringService.utilRecordUsageObject(trackedUsage, actor, modelDetails.id, costOverwrites);
|
||||
const legacyCostCalculator = OpenAIUtil.create_usage_calculator({
|
||||
model_details: modelDetails,
|
||||
});
|
||||
return legacyCostCalculator({ usage });
|
||||
},
|
||||
},
|
||||
};
|
||||
stream,
|
||||
completion,
|
||||
});
|
||||
}
|
||||
|
||||
/**
|
||||
* Retrieves available AI models and their specifications
|
||||
@@ -173,19 +171,21 @@ class OpenRouterService extends BaseService {
|
||||
* - name: Human readable model name
|
||||
* - context: Maximum context window size
|
||||
* - cost: Pricing information object with currency and rates
|
||||
* @private
|
||||
*/
|
||||
async models_(rawPriceKeys = false) {
|
||||
const axios = this.require('axios');
|
||||
|
||||
async models (rawPriceKeys = false) {
|
||||
let models = this.modules.kv.get(`${this.kvkey}:models`);
|
||||
if ( !models ) {
|
||||
const resp = await axios.request({
|
||||
method: 'GET',
|
||||
url: this.api_base_url + '/models',
|
||||
});
|
||||
models = resp.data.data;
|
||||
this.modules.kv.set(`${this.kvkey}:models`, models);
|
||||
if ( ! models ) {
|
||||
try {
|
||||
const resp = await axios.request({
|
||||
method: 'GET',
|
||||
url: `${this.api_base_url}/models`,
|
||||
});
|
||||
|
||||
models = resp.data.data;
|
||||
this.modules.kv.set(`${this.kvkey}:models`, models);
|
||||
} catch (e) {
|
||||
console.log(e);
|
||||
}
|
||||
}
|
||||
const coerced_models = [];
|
||||
for ( const model of models ) {
|
||||
@@ -194,8 +194,8 @@ class OpenRouterService extends BaseService {
|
||||
output: Math.round(model.pricing.completion * 1_000_000 * 100),
|
||||
};
|
||||
coerced_models.push({
|
||||
id: 'openrouter:' + model.id,
|
||||
name: model.name + ' (OpenRouter)',
|
||||
id: `openrouter:${ model.id}`,
|
||||
name: `${model.name } (OpenRouter)`,
|
||||
max_tokens: model.top_provider.max_completion_tokens,
|
||||
cost: {
|
||||
currency: 'usd-cents',
|
||||
|
||||
@@ -24,7 +24,7 @@ const process_input_messages = async (messages) => {
|
||||
const content_block = content[i];
|
||||
|
||||
if ( content_block.type === 'tool_use' ) {
|
||||
if ( !msg.tool_calls ) {
|
||||
if ( ! msg.tool_calls ) {
|
||||
msg.tool_calls = [];
|
||||
is_tool_call = true;
|
||||
}
|
||||
@@ -108,12 +108,6 @@ const create_chat_stream_handler = ({
|
||||
let last_usage = null;
|
||||
for await ( let chunk of completion ) {
|
||||
chunk = deviations.chunk_but_like_actually(chunk);
|
||||
if ( process.env.DEBUG ) {
|
||||
const delta = chunk?.choices?.[0]?.delta;
|
||||
console.log(`AI CHUNK`,
|
||||
chunk,
|
||||
delta && JSON.stringify(delta));
|
||||
}
|
||||
const chunk_usage = deviations.index_usage_from_stream_chunk(chunk);
|
||||
if ( chunk_usage ) last_usage = chunk_usage;
|
||||
if ( chunk.choices.length < 1 ) continue;
|
||||
@@ -127,7 +121,7 @@ const create_chat_stream_handler = ({
|
||||
// A: For now, reasoning_content and content never appear together, but I’m not sure if they’ll always be mutually exclusive.
|
||||
}
|
||||
|
||||
if ( choice.delta.content ){
|
||||
if ( choice.delta.content ) {
|
||||
if ( mode === 'tool' ) {
|
||||
toolblock.end();
|
||||
mode = 'text';
|
||||
@@ -205,7 +199,7 @@ const handle_completion_output = async ({
|
||||
if ( finally_fn ) await finally_fn();
|
||||
|
||||
const is_empty = completion.choices?.[0]?.message?.content?.trim() === '';
|
||||
if ( is_empty && ! completion.choices?.[0]?.message?.tool_calls ) {
|
||||
if ( is_empty && !completion.choices?.[0]?.message?.tool_calls ) {
|
||||
// GPT refuses to generate an empty response if you ask it to,
|
||||
// so this will probably only happen on an error condition.
|
||||
throw new Error('an empty response was generated');
|
||||
|
||||
@@ -353,7 +353,7 @@ export class MeteringService {
|
||||
}
|
||||
|
||||
async getActorCurrentMonthUsageDetails (actor: Actor) {
|
||||
if ( !actor.type?.user?.uuid ) {
|
||||
if ( ! actor.type?.user?.uuid ) {
|
||||
throw new Error('Actor must be a user to get usage details');
|
||||
}
|
||||
// batch get actor usage, per app usage, and actor app totals for the month
|
||||
@@ -375,7 +375,7 @@ export class MeteringService {
|
||||
filteredAppTotals[appKey] = appUsage;
|
||||
} else {
|
||||
Object.entries(appUsage).forEach(([usageKind, amount]) => {
|
||||
if ( !othersTotal[usageKind as keyof AppTotals] ) {
|
||||
if ( ! othersTotal[usageKind as keyof AppTotals] ) {
|
||||
othersTotal[usageKind as keyof AppTotals] = 0;
|
||||
}
|
||||
othersTotal[usageKind as keyof AppTotals] += amount;
|
||||
@@ -398,7 +398,7 @@ export class MeteringService {
|
||||
}
|
||||
|
||||
async getActorCurrentMonthAppUsageDetails (actor: Actor, appId?: string) {
|
||||
if ( !actor.type?.user?.uuid ) {
|
||||
if ( ! actor.type?.user?.uuid ) {
|
||||
throw new Error('Actor must be a user to get usage details');
|
||||
}
|
||||
appId = appId || actor.type?.app?.uid || GLOBAL_APP_KEY;
|
||||
@@ -453,7 +453,7 @@ export class MeteringService {
|
||||
|
||||
async getActorSubscription (actor: Actor): Promise<(typeof SUB_POLICIES)[number]> {
|
||||
// TODO DS: maybe allow non-user actors to have subscriptions eventually
|
||||
if ( !actor.type?.user.uuid ) {
|
||||
if ( ! actor.type?.user.uuid ) {
|
||||
throw new Error('Actor must be a user to get policy');
|
||||
}
|
||||
|
||||
@@ -476,7 +476,7 @@ export class MeteringService {
|
||||
}
|
||||
|
||||
async getActorAddons (actor: Actor) {
|
||||
if ( !actor.type?.user?.uuid ) {
|
||||
if ( ! actor.type?.user?.uuid ) {
|
||||
throw new Error('Actor must be a user to get policy addons');
|
||||
}
|
||||
const key = `${POLICY_PREFIX}:actor:${actor.type.user?.uuid}:addons`;
|
||||
@@ -487,7 +487,7 @@ export class MeteringService {
|
||||
}
|
||||
|
||||
async getActorAppUsage (actor: Actor, appId: string) {
|
||||
if ( !actor.type?.user?.uuid ) {
|
||||
if ( ! actor.type?.user?.uuid ) {
|
||||
throw new Error('Actor must be a user to get app usage');
|
||||
}
|
||||
|
||||
@@ -522,7 +522,7 @@ export class MeteringService {
|
||||
aggregatedUsage.total += total || 0;
|
||||
|
||||
Object.entries((usage || {}) as Record<string, UsageRecord>).forEach(([usageKind, record]) => {
|
||||
if ( !aggregatedUsage[usageKind] ) {
|
||||
if ( ! aggregatedUsage[usageKind] ) {
|
||||
aggregatedUsage[usageKind] = { cost: 0, units: 0, count: 0 } as UsageRecord;
|
||||
}
|
||||
const aggregatedRecord = aggregatedUsage[usageKind] as UsageRecord;
|
||||
@@ -535,8 +535,8 @@ export class MeteringService {
|
||||
});
|
||||
}
|
||||
|
||||
async updateAddonCredit (userId:string, tokenAmount: number) {
|
||||
if ( !userId ) {
|
||||
async updateAddonCredit (userId: string, tokenAmount: number) {
|
||||
if ( ! userId ) {
|
||||
throw new Error('User needed to update extra credits');
|
||||
}
|
||||
const key = `${POLICY_PREFIX}:actor:${userId}:addons`;
|
||||
|
||||
@@ -1,15 +1,71 @@
|
||||
export const OPENROUTER_COST_MAP = {
|
||||
|
||||
'openrouter:google/gemini-3-pro-preview:prompt': 200,
|
||||
'openrouter:google/gemini-3-pro-preview:completion': 1200,
|
||||
'openrouter:google/gemini-3-pro-preview:image': 825600,
|
||||
'openrouter:google/gemini-3-pro-preview:input_cache_read': 20,
|
||||
'openrouter:google/gemini-3-pro-preview:input_cache_write': 238,
|
||||
'openrouter:deepcogito/cogito-v2.1-671b:prompt': 125,
|
||||
'openrouter:deepcogito/cogito-v2.1-671b:completion': 125,
|
||||
'openrouter:openai/gpt-5.1:prompt': 125,
|
||||
'openrouter:openai/gpt-5.1:completion': 1000,
|
||||
'openrouter:openai/gpt-5.1:web_search': 1000000,
|
||||
'openrouter:openai/gpt-5.1:input_cache_read': 12,
|
||||
'openrouter:openai/gpt-5.1-chat:prompt': 125,
|
||||
'openrouter:openai/gpt-5.1-chat:completion': 1000,
|
||||
'openrouter:openai/gpt-5.1-chat:web_search': 1000000,
|
||||
'openrouter:openai/gpt-5.1-chat:input_cache_read': 12,
|
||||
'openrouter:openai/gpt-5.1-codex:prompt': 125,
|
||||
'openrouter:openai/gpt-5.1-codex:completion': 1000,
|
||||
'openrouter:openai/gpt-5.1-codex:input_cache_read': 12,
|
||||
'openrouter:openai/gpt-5.1-codex-mini:prompt': 25,
|
||||
'openrouter:openai/gpt-5.1-codex-mini:completion': 200,
|
||||
'openrouter:openai/gpt-5.1-codex-mini:input_cache_read': 3,
|
||||
'openrouter:moonshotai/kimi-linear-48b-a3b-instruct:prompt': 50,
|
||||
'openrouter:moonshotai/kimi-linear-48b-a3b-instruct:completion': 60,
|
||||
'openrouter:moonshotai/kimi-k2-thinking:prompt': 45,
|
||||
'openrouter:moonshotai/kimi-k2-thinking:completion': 235,
|
||||
'openrouter:amazon/nova-premier-v1:prompt': 250,
|
||||
'openrouter:amazon/nova-premier-v1:completion': 1250,
|
||||
'openrouter:amazon/nova-premier-v1:input_cache_read': 63,
|
||||
'openrouter:perplexity/sonar-pro-search:prompt': 300,
|
||||
'openrouter:perplexity/sonar-pro-search:completion': 1500,
|
||||
'openrouter:perplexity/sonar-pro-search:request': 1800000,
|
||||
'openrouter:mistralai/voxtral-small-24b-2507:prompt': 10,
|
||||
'openrouter:mistralai/voxtral-small-24b-2507:completion': 30,
|
||||
'openrouter:mistralai/voxtral-small-24b-2507:audio': 10000,
|
||||
'openrouter:openai/gpt-oss-safeguard-20b:prompt': 7,
|
||||
'openrouter:openai/gpt-oss-safeguard-20b:completion': 30,
|
||||
'openrouter:openai/gpt-oss-safeguard-20b:input_cache_read': 4,
|
||||
'openrouter:nvidia/nemotron-nano-12b-v2-vl:prompt': 20,
|
||||
'openrouter:nvidia/nemotron-nano-12b-v2-vl:completion': 60,
|
||||
'openrouter:minimax/minimax-m2:prompt': 26,
|
||||
'openrouter:minimax/minimax-m2:completion': 102,
|
||||
'openrouter:liquid/lfm2-8b-a1b:prompt': 5,
|
||||
'openrouter:liquid/lfm2-8b-a1b:completion': 10,
|
||||
'openrouter:liquid/lfm-2.2-6b:prompt': 5,
|
||||
'openrouter:liquid/lfm-2.2-6b:completion': 10,
|
||||
'openrouter:ibm-granite/granite-4.0-h-micro:prompt': 2,
|
||||
'openrouter:ibm-granite/granite-4.0-h-micro:completion': 11,
|
||||
'openrouter:deepcogito/cogito-v2-preview-llama-405b:prompt': 350,
|
||||
'openrouter:deepcogito/cogito-v2-preview-llama-405b:completion': 350,
|
||||
'openrouter:openai/gpt-5-image-mini:prompt': 250,
|
||||
'openrouter:openai/gpt-5-image-mini:completion': 200,
|
||||
'openrouter:openai/gpt-5-image-mini:image': 250,
|
||||
'openrouter:openai/gpt-5-image-mini:web_search': 1000000,
|
||||
'openrouter:openai/gpt-5-image-mini:input_cache_read': 25,
|
||||
'openrouter:anthropic/claude-haiku-4.5:prompt': 100,
|
||||
'openrouter:anthropic/claude-haiku-4.5:completion': 500,
|
||||
'openrouter:anthropic/claude-haiku-4.5:input_cache_read': 10,
|
||||
'openrouter:anthropic/claude-haiku-4.5:input_cache_write': 125,
|
||||
'openrouter:qwen/qwen3-vl-8b-thinking:prompt': 18,
|
||||
'openrouter:qwen/qwen3-vl-8b-thinking:completion': 210,
|
||||
'openrouter:qwen/qwen3-vl-8b-instruct:prompt': 18,
|
||||
'openrouter:qwen/qwen3-vl-8b-instruct:completion': 69,
|
||||
'openrouter:inclusionai/ling-1t:prompt': 100,
|
||||
'openrouter:inclusionai/ling-1t:completion': 300,
|
||||
'openrouter:qwen/qwen3-vl-8b-instruct:prompt': 8,
|
||||
'openrouter:qwen/qwen3-vl-8b-instruct:completion': 50,
|
||||
'openrouter:openai/gpt-5-image:prompt': 1000,
|
||||
'openrouter:openai/gpt-5-image:completion': 1000,
|
||||
'openrouter:openai/gpt-5-image:image': 1000,
|
||||
'openrouter:openai/gpt-5-image:web_search': 1000000,
|
||||
'openrouter:openai/gpt-5-image:input_cache_read': 125,
|
||||
'openrouter:openai/o3-deep-research:prompt': 1000,
|
||||
'openrouter:openai/o3-deep-research:completion': 4000,
|
||||
'openrouter:openai/o3-deep-research:image': 765000,
|
||||
@@ -27,16 +83,21 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:google/gemini-2.5-flash-image:prompt': 30,
|
||||
'openrouter:google/gemini-2.5-flash-image:completion': 250,
|
||||
'openrouter:google/gemini-2.5-flash-image:image': 123800,
|
||||
'openrouter:qwen/qwen3-vl-30b-a3b-thinking:prompt': 29,
|
||||
'openrouter:qwen/qwen3-vl-30b-a3b-thinking:prompt': 20,
|
||||
'openrouter:qwen/qwen3-vl-30b-a3b-thinking:completion': 100,
|
||||
'openrouter:qwen/qwen3-vl-30b-a3b-instruct:prompt': 29,
|
||||
'openrouter:qwen/qwen3-vl-30b-a3b-instruct:completion': 99,
|
||||
'openrouter:qwen/qwen3-vl-30b-a3b-instruct:prompt': 15,
|
||||
'openrouter:qwen/qwen3-vl-30b-a3b-instruct:completion': 60,
|
||||
'openrouter:openai/gpt-5-pro:prompt': 1500,
|
||||
'openrouter:openai/gpt-5-pro:completion': 12000,
|
||||
'openrouter:z-ai/glm-4.6:prompt': 50,
|
||||
'openrouter:openai/gpt-5-pro:web_search': 1000000,
|
||||
'openrouter:z-ai/glm-4.6:prompt': 40,
|
||||
'openrouter:z-ai/glm-4.6:completion': 175,
|
||||
'openrouter:z-ai/glm-4.6:exacto:prompt': 45,
|
||||
'openrouter:z-ai/glm-4.6:exacto:completion': 190,
|
||||
'openrouter:anthropic/claude-sonnet-4.5:prompt': 300,
|
||||
'openrouter:anthropic/claude-sonnet-4.5:completion': 1500,
|
||||
'openrouter:anthropic/claude-sonnet-4.5:input_cache_read': 30,
|
||||
'openrouter:anthropic/claude-sonnet-4.5:input_cache_write': 375,
|
||||
'openrouter:deepseek/deepseek-v3.2-exp:prompt': 27,
|
||||
'openrouter:deepseek/deepseek-v3.2-exp:completion': 40,
|
||||
'openrouter:thedrummer/cydonia-24b-v4.1:prompt': 30,
|
||||
@@ -46,14 +107,15 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:google/gemini-2.5-flash-preview-09-2025:prompt': 30,
|
||||
'openrouter:google/gemini-2.5-flash-preview-09-2025:completion': 250,
|
||||
'openrouter:google/gemini-2.5-flash-preview-09-2025:image': 123800,
|
||||
'openrouter:google/gemini-2.5-flash-preview-09-2025:audio': 100,
|
||||
'openrouter:google/gemini-2.5-flash-preview-09-2025:input_cache_read': 7,
|
||||
'openrouter:google/gemini-2.5-flash-preview-09-2025:input_cache_write': 38,
|
||||
'openrouter:google/gemini-2.5-flash-lite-preview-09-2025:prompt': 10,
|
||||
'openrouter:google/gemini-2.5-flash-lite-preview-09-2025:completion': 40,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-thinking:prompt': 45,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-thinking:completion': 350,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-instruct:prompt': 30,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-instruct:completion': 120,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-thinking:prompt': 30,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-thinking:completion': 120,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-instruct:prompt': 21,
|
||||
'openrouter:qwen/qwen3-vl-235b-a22b-instruct:completion': 190,
|
||||
'openrouter:qwen/qwen3-max:prompt': 120,
|
||||
'openrouter:qwen/qwen3-max:completion': 600,
|
||||
'openrouter:qwen/qwen3-max:input_cache_read': 24,
|
||||
@@ -65,6 +127,8 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:openai/gpt-5-codex:input_cache_read': 12,
|
||||
'openrouter:deepseek/deepseek-v3.1-terminus:prompt': 23,
|
||||
'openrouter:deepseek/deepseek-v3.1-terminus:completion': 90,
|
||||
'openrouter:deepseek/deepseek-v3.1-terminus:exacto:prompt': 27,
|
||||
'openrouter:deepseek/deepseek-v3.1-terminus:exacto:completion': 100,
|
||||
'openrouter:x-ai/grok-4-fast:prompt': 20,
|
||||
'openrouter:x-ai/grok-4-fast:completion': 50,
|
||||
'openrouter:x-ai/grok-4-fast:input_cache_read': 5,
|
||||
@@ -77,7 +141,7 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:arcee-ai/afm-4.5b:completion': 15,
|
||||
'openrouter:opengvlab/internvl3-78b:prompt': 7,
|
||||
'openrouter:opengvlab/internvl3-78b:completion': 26,
|
||||
'openrouter:qwen/qwen3-next-80b-a3b-thinking:prompt': 14,
|
||||
'openrouter:qwen/qwen3-next-80b-a3b-thinking:prompt': 15,
|
||||
'openrouter:qwen/qwen3-next-80b-a3b-thinking:completion': 120,
|
||||
'openrouter:qwen/qwen3-next-80b-a3b-instruct:prompt': 10,
|
||||
'openrouter:qwen/qwen3-next-80b-a3b-instruct:completion': 80,
|
||||
@@ -91,14 +155,18 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:nvidia/nemotron-nano-9b-v2:completion': 16,
|
||||
'openrouter:moonshotai/kimi-k2-0905:prompt': 39,
|
||||
'openrouter:moonshotai/kimi-k2-0905:completion': 190,
|
||||
'openrouter:moonshotai/kimi-k2-0905:exacto:prompt': 60,
|
||||
'openrouter:moonshotai/kimi-k2-0905:exacto:completion': 250,
|
||||
'openrouter:deepcogito/cogito-v2-preview-llama-70b:prompt': 88,
|
||||
'openrouter:deepcogito/cogito-v2-preview-llama-70b:completion': 88,
|
||||
'openrouter:deepcogito/cogito-v2-preview-llama-109b-moe:prompt': 18,
|
||||
'openrouter:deepcogito/cogito-v2-preview-llama-109b-moe:completion': 59,
|
||||
'openrouter:deepcogito/cogito-v2-preview-deepseek-671b:prompt': 125,
|
||||
'openrouter:deepcogito/cogito-v2-preview-deepseek-671b:completion': 125,
|
||||
'openrouter:stepfun-ai/step3:prompt': 57,
|
||||
'openrouter:stepfun-ai/step3:completion': 142,
|
||||
'openrouter:qwen/qwen3-30b-a3b-thinking-2507:prompt': 8,
|
||||
'openrouter:qwen/qwen3-30b-a3b-thinking-2507:completion': 29,
|
||||
'openrouter:qwen/qwen3-30b-a3b-thinking-2507:prompt': 5,
|
||||
'openrouter:qwen/qwen3-30b-a3b-thinking-2507:completion': 34,
|
||||
'openrouter:x-ai/grok-code-fast-1:prompt': 20,
|
||||
'openrouter:x-ai/grok-code-fast-1:completion': 150,
|
||||
'openrouter:x-ai/grok-code-fast-1:input_cache_read': 2,
|
||||
@@ -129,20 +197,8 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:ai21/jamba-large-1.7:completion': 800,
|
||||
'openrouter:openai/gpt-5-chat:prompt': 125,
|
||||
'openrouter:openai/gpt-5-chat:completion': 1000,
|
||||
'openrouter:openai/gpt-5-chat:web_search': 1000000,
|
||||
'openrouter:openai/gpt-5-chat:input_cache_read': 12,
|
||||
'openrouter:openai/gpt-5.1:prompt': 125,
|
||||
'openrouter:openai/gpt-5.1:completion': 1000,
|
||||
'openrouter:openai/gpt-5.1:web_search': 1000000,
|
||||
'openrouter:openai/gpt-5.1:input_cache_read': 12,
|
||||
'openrouter:openai/gpt-5.1-codex:prompt': 125,
|
||||
'openrouter:openai/gpt-5.1-codex:completion': 1000,
|
||||
'openrouter:openai/gpt-5.1-codex:input_cache_read': 12,
|
||||
'openrouter:openai/gpt-5.1-codex-mini:prompt': 25,
|
||||
'openrouter:openai/gpt-5.1-codex-mini:completion': 200,
|
||||
'openrouter:openai/gpt-5.1-codex-mini:input_cache_read': 3,
|
||||
'openrouter:openai/gpt-5.1-chat-latest:prompt': 125,
|
||||
'openrouter:openai/gpt-5.1-chat-latest:completion': 1000,
|
||||
'openrouter:openai/gpt-5.1-chat-latest:input_cache_read': 12,
|
||||
'openrouter:openai/gpt-5:prompt': 125,
|
||||
'openrouter:openai/gpt-5:completion': 1000,
|
||||
'openrouter:openai/gpt-5:web_search': 1000000,
|
||||
@@ -157,6 +213,8 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:openai/gpt-5-nano:input_cache_read': 1,
|
||||
'openrouter:openai/gpt-oss-120b:prompt': 4,
|
||||
'openrouter:openai/gpt-oss-120b:completion': 40,
|
||||
'openrouter:openai/gpt-oss-120b:exacto:prompt': 5,
|
||||
'openrouter:openai/gpt-oss-120b:exacto:completion': 24,
|
||||
'openrouter:openai/gpt-oss-20b:prompt': 3,
|
||||
'openrouter:openai/gpt-oss-20b:completion': 14,
|
||||
'openrouter:anthropic/claude-opus-4.1:prompt': 1500,
|
||||
@@ -171,15 +229,17 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:qwen/qwen3-30b-a3b-instruct-2507:prompt': 8,
|
||||
'openrouter:qwen/qwen3-30b-a3b-instruct-2507:completion': 33,
|
||||
'openrouter:z-ai/glm-4.5:prompt': 35,
|
||||
'openrouter:z-ai/glm-4.5:completion': 155,
|
||||
'openrouter:z-ai/glm-4.5-air:prompt': 14,
|
||||
'openrouter:z-ai/glm-4.5-air:completion': 86,
|
||||
'openrouter:z-ai/glm-4.5:completion': 150,
|
||||
'openrouter:z-ai/glm-4.5-air:prompt': 13,
|
||||
'openrouter:z-ai/glm-4.5-air:completion': 85,
|
||||
'openrouter:qwen/qwen3-235b-a22b-thinking-2507:prompt': 11,
|
||||
'openrouter:qwen/qwen3-235b-a22b-thinking-2507:completion': 60,
|
||||
'openrouter:z-ai/glm-4-32b:prompt': 10,
|
||||
'openrouter:z-ai/glm-4-32b:completion': 10,
|
||||
'openrouter:qwen/qwen3-coder:prompt': 22,
|
||||
'openrouter:qwen/qwen3-coder:completion': 95,
|
||||
'openrouter:qwen/qwen3-coder:exacto:prompt': 38,
|
||||
'openrouter:qwen/qwen3-coder:exacto:completion': 153,
|
||||
'openrouter:bytedance/ui-tars-1.5-7b:prompt': 10,
|
||||
'openrouter:bytedance/ui-tars-1.5-7b:completion': 20,
|
||||
'openrouter:google/gemini-2.5-flash-lite:prompt': 10,
|
||||
@@ -190,8 +250,8 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:qwen/qwen3-235b-a22b-2507:completion': 55,
|
||||
'openrouter:switchpoint/router:prompt': 85,
|
||||
'openrouter:switchpoint/router:completion': 340,
|
||||
'openrouter:moonshotai/kimi-k2:prompt': 14,
|
||||
'openrouter:moonshotai/kimi-k2:completion': 249,
|
||||
'openrouter:moonshotai/kimi-k2:prompt': 50,
|
||||
'openrouter:moonshotai/kimi-k2:completion': 240,
|
||||
'openrouter:thudm/glm-4.1v-9b-thinking:prompt': 4,
|
||||
'openrouter:thudm/glm-4.1v-9b-thinking:completion': 14,
|
||||
'openrouter:mistralai/devstral-medium:prompt': 40,
|
||||
@@ -201,8 +261,8 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:x-ai/grok-4:prompt': 300,
|
||||
'openrouter:x-ai/grok-4:completion': 1500,
|
||||
'openrouter:x-ai/grok-4:input_cache_read': 75,
|
||||
'openrouter:tencent/hunyuan-a13b-instruct:prompt': 3,
|
||||
'openrouter:tencent/hunyuan-a13b-instruct:completion': 3,
|
||||
'openrouter:tencent/hunyuan-a13b-instruct:prompt': 14,
|
||||
'openrouter:tencent/hunyuan-a13b-instruct:completion': 57,
|
||||
'openrouter:tngtech/deepseek-r1t2-chimera:prompt': 30,
|
||||
'openrouter:tngtech/deepseek-r1t2-chimera:completion': 120,
|
||||
'openrouter:morph/morph-v3-large:prompt': 90,
|
||||
@@ -221,11 +281,6 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:mistralai/mistral-small-3.2-24b-instruct:completion': 18,
|
||||
'openrouter:minimax/minimax-m1:prompt': 40,
|
||||
'openrouter:minimax/minimax-m1:completion': 220,
|
||||
'openrouter:google/gemini-2.5-flash-lite-preview-06-17:prompt': 10,
|
||||
'openrouter:google/gemini-2.5-flash-lite-preview-06-17:completion': 40,
|
||||
'openrouter:google/gemini-2.5-flash-lite-preview-06-17:audio': 30,
|
||||
'openrouter:google/gemini-2.5-flash-lite-preview-06-17:input_cache_read': 3,
|
||||
'openrouter:google/gemini-2.5-flash-lite-preview-06-17:input_cache_write': 18,
|
||||
'openrouter:google/gemini-2.5-flash:prompt': 30,
|
||||
'openrouter:google/gemini-2.5-flash:completion': 250,
|
||||
'openrouter:google/gemini-2.5-flash:image': 123800,
|
||||
@@ -250,19 +305,19 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:x-ai/grok-3:input_cache_read': 75,
|
||||
'openrouter:mistralai/magistral-small-2506:prompt': 50,
|
||||
'openrouter:mistralai/magistral-small-2506:completion': 150,
|
||||
'openrouter:mistralai/magistral-medium-2506:prompt': 200,
|
||||
'openrouter:mistralai/magistral-medium-2506:completion': 500,
|
||||
'openrouter:mistralai/magistral-medium-2506:thinking:prompt': 200,
|
||||
'openrouter:mistralai/magistral-medium-2506:thinking:completion': 500,
|
||||
'openrouter:mistralai/magistral-medium-2506:prompt': 200,
|
||||
'openrouter:mistralai/magistral-medium-2506:completion': 500,
|
||||
'openrouter:google/gemini-2.5-pro-preview:prompt': 125,
|
||||
'openrouter:google/gemini-2.5-pro-preview:completion': 1000,
|
||||
'openrouter:google/gemini-2.5-pro-preview:image': 516000,
|
||||
'openrouter:google/gemini-2.5-pro-preview:input_cache_read': 31,
|
||||
'openrouter:google/gemini-2.5-pro-preview:input_cache_write': 163,
|
||||
'openrouter:deepseek/deepseek-r1-0528-qwen3-8b:prompt': 3,
|
||||
'openrouter:deepseek/deepseek-r1-0528-qwen3-8b:completion': 11,
|
||||
'openrouter:deepseek/deepseek-r1-0528:prompt': 40,
|
||||
'openrouter:deepseek/deepseek-r1-0528:completion': 175,
|
||||
'openrouter:deepseek/deepseek-r1-0528-qwen3-8b:prompt': 2,
|
||||
'openrouter:deepseek/deepseek-r1-0528-qwen3-8b:completion': 10,
|
||||
'openrouter:deepseek/deepseek-r1-0528:prompt': 20,
|
||||
'openrouter:deepseek/deepseek-r1-0528:completion': 450,
|
||||
'openrouter:anthropic/claude-opus-4:prompt': 1500,
|
||||
'openrouter:anthropic/claude-opus-4:completion': 7500,
|
||||
'openrouter:anthropic/claude-opus-4:image': 2400000,
|
||||
@@ -273,8 +328,8 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:anthropic/claude-sonnet-4:image': 480000,
|
||||
'openrouter:anthropic/claude-sonnet-4:input_cache_read': 30,
|
||||
'openrouter:anthropic/claude-sonnet-4:input_cache_write': 375,
|
||||
'openrouter:mistralai/devstral-small-2505:prompt': 5,
|
||||
'openrouter:mistralai/devstral-small-2505:completion': 22,
|
||||
'openrouter:mistralai/devstral-small-2505:prompt': 6,
|
||||
'openrouter:mistralai/devstral-small-2505:completion': 12,
|
||||
'openrouter:google/gemma-3n-e4b-it:prompt': 2,
|
||||
'openrouter:google/gemma-3n-e4b-it:completion': 4,
|
||||
'openrouter:openai/codex-mini:prompt': 150,
|
||||
@@ -319,8 +374,6 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:tngtech/deepseek-r1t-chimera:completion': 120,
|
||||
'openrouter:microsoft/mai-ds-r1:prompt': 30,
|
||||
'openrouter:microsoft/mai-ds-r1:completion': 120,
|
||||
'openrouter:thudm/glm-z1-32b:prompt': 5,
|
||||
'openrouter:thudm/glm-z1-32b:completion': 22,
|
||||
'openrouter:openai/o4-mini-high:prompt': 110,
|
||||
'openrouter:openai/o4-mini-high:completion': 440,
|
||||
'openrouter:openai/o4-mini-high:image': 84150,
|
||||
@@ -336,8 +389,6 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:openai/o4-mini:image': 84150,
|
||||
'openrouter:openai/o4-mini:web_search': 1000000,
|
||||
'openrouter:openai/o4-mini:input_cache_read': 28,
|
||||
'openrouter:shisa-ai/shisa-v2-llama3.3-70b:prompt': 5,
|
||||
'openrouter:shisa-ai/shisa-v2-llama3.3-70b:completion': 22,
|
||||
'openrouter:qwen/qwen2.5-coder-7b-instruct:prompt': 3,
|
||||
'openrouter:qwen/qwen2.5-coder-7b-instruct:completion': 9,
|
||||
'openrouter:openai/gpt-4.1:prompt': 200,
|
||||
@@ -358,8 +409,6 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:alfredpros/codellama-7b-instruct-solidity:completion': 120,
|
||||
'openrouter:arliai/qwq-32b-arliai-rpr-v1:prompt': 3,
|
||||
'openrouter:arliai/qwq-32b-arliai-rpr-v1:completion': 11,
|
||||
'openrouter:agentica-org/deepcoder-14b-preview:prompt': 1,
|
||||
'openrouter:agentica-org/deepcoder-14b-preview:completion': 1,
|
||||
'openrouter:x-ai/grok-3-mini-beta:prompt': 30,
|
||||
'openrouter:x-ai/grok-3-mini-beta:completion': 50,
|
||||
'openrouter:x-ai/grok-3-mini-beta:input_cache_read': 7,
|
||||
@@ -374,8 +423,6 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:meta-llama/llama-4-scout:prompt': 8,
|
||||
'openrouter:meta-llama/llama-4-scout:completion': 30,
|
||||
'openrouter:meta-llama/llama-4-scout:image': 33420,
|
||||
'openrouter:allenai/molmo-7b-d:prompt': 10,
|
||||
'openrouter:allenai/molmo-7b-d:completion': 20,
|
||||
'openrouter:qwen/qwen2.5-vl-32b-instruct:prompt': 5,
|
||||
'openrouter:qwen/qwen2.5-vl-32b-instruct:completion': 22,
|
||||
'openrouter:deepseek/deepseek-chat-v3-0324:prompt': 24,
|
||||
@@ -384,7 +431,7 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:openai/o1-pro:completion': 60000,
|
||||
'openrouter:openai/o1-pro:image': 21675000,
|
||||
'openrouter:mistralai/mistral-small-3.1-24b-instruct:prompt': 5,
|
||||
'openrouter:mistralai/mistral-small-3.1-24b-instruct:completion': 10,
|
||||
'openrouter:mistralai/mistral-small-3.1-24b-instruct:completion': 22,
|
||||
'openrouter:allenai/olmo-2-0325-32b-instruct:prompt': 20,
|
||||
'openrouter:allenai/olmo-2-0325-32b-instruct:completion': 35,
|
||||
'openrouter:google/gemma-3-4b-it:prompt': 2,
|
||||
@@ -401,11 +448,10 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:openai/gpt-4o-search-preview:completion': 1000,
|
||||
'openrouter:openai/gpt-4o-search-preview:request': 3500000,
|
||||
'openrouter:openai/gpt-4o-search-preview:image': 361300,
|
||||
'openrouter:google/gemma-3-27b-it:prompt': 9,
|
||||
'openrouter:google/gemma-3-27b-it:completion': 16,
|
||||
'openrouter:google/gemma-3-27b-it:image': 2560,
|
||||
'openrouter:thedrummer/skyfall-36b-v2:prompt': 8,
|
||||
'openrouter:thedrummer/skyfall-36b-v2:completion': 33,
|
||||
'openrouter:google/gemma-3-27b-it:prompt': 7,
|
||||
'openrouter:google/gemma-3-27b-it:completion': 50,
|
||||
'openrouter:thedrummer/skyfall-36b-v2:prompt': 50,
|
||||
'openrouter:thedrummer/skyfall-36b-v2:completion': 80,
|
||||
'openrouter:microsoft/phi-4-multimodal-instruct:prompt': 5,
|
||||
'openrouter:microsoft/phi-4-multimodal-instruct:completion': 10,
|
||||
'openrouter:microsoft/phi-4-multimodal-instruct:image': 17685,
|
||||
@@ -421,24 +467,20 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:perplexity/sonar-deep-research:internal_reasoning': 300,
|
||||
'openrouter:qwen/qwq-32b:prompt': 15,
|
||||
'openrouter:qwen/qwq-32b:completion': 40,
|
||||
'openrouter:nousresearch/deephermes-3-llama-3-8b-preview:prompt': 3,
|
||||
'openrouter:nousresearch/deephermes-3-llama-3-8b-preview:completion': 11,
|
||||
'openrouter:google/gemini-2.0-flash-lite-001:prompt': 7,
|
||||
'openrouter:google/gemini-2.0-flash-lite-001:completion': 30,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:prompt': 300,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:completion': 1500,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:image': 480000,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:input_cache_read': 30,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:input_cache_write': 375,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:thinking:prompt': 300,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:thinking:completion': 1500,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:thinking:image': 480000,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:thinking:input_cache_read': 30,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:thinking:input_cache_write': 375,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:prompt': 300,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:completion': 1500,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:image': 480000,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:input_cache_read': 30,
|
||||
'openrouter:anthropic/claude-3.7-sonnet:input_cache_write': 375,
|
||||
'openrouter:mistralai/mistral-saba:prompt': 20,
|
||||
'openrouter:mistralai/mistral-saba:completion': 60,
|
||||
'openrouter:cognitivecomputations/dolphin3.0-mistral-24b:prompt': 4,
|
||||
'openrouter:cognitivecomputations/dolphin3.0-mistral-24b:completion': 17,
|
||||
'openrouter:meta-llama/llama-guard-3-8b:prompt': 2,
|
||||
'openrouter:meta-llama/llama-guard-3-8b:completion': 6,
|
||||
'openrouter:openai/o3-mini-high:prompt': 110,
|
||||
@@ -488,14 +530,10 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:perplexity/sonar:prompt': 100,
|
||||
'openrouter:perplexity/sonar:completion': 100,
|
||||
'openrouter:perplexity/sonar:request': 500000,
|
||||
'openrouter:liquid/lfm-7b:prompt': 1,
|
||||
'openrouter:liquid/lfm-7b:completion': 1,
|
||||
'openrouter:liquid/lfm-3b:prompt': 2,
|
||||
'openrouter:liquid/lfm-3b:completion': 2,
|
||||
'openrouter:deepseek/deepseek-r1-distill-llama-70b:prompt': 3,
|
||||
'openrouter:deepseek/deepseek-r1-distill-llama-70b:completion': 13,
|
||||
'openrouter:deepseek/deepseek-r1:prompt': 40,
|
||||
'openrouter:deepseek/deepseek-r1:completion': 200,
|
||||
'openrouter:deepseek/deepseek-r1:prompt': 30,
|
||||
'openrouter:deepseek/deepseek-r1:completion': 120,
|
||||
'openrouter:minimax/minimax-01:prompt': 20,
|
||||
'openrouter:minimax/minimax-01:completion': 110,
|
||||
'openrouter:mistralai/codestral-2501:prompt': 30,
|
||||
@@ -505,7 +543,7 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:sao10k/l3.1-70b-hanami-x1:prompt': 300,
|
||||
'openrouter:sao10k/l3.1-70b-hanami-x1:completion': 300,
|
||||
'openrouter:deepseek/deepseek-chat:prompt': 30,
|
||||
'openrouter:deepseek/deepseek-chat:completion': 85,
|
||||
'openrouter:deepseek/deepseek-chat:completion': 120,
|
||||
'openrouter:sao10k/l3.3-euryale-70b:prompt': 65,
|
||||
'openrouter:sao10k/l3.3-euryale-70b:completion': 75,
|
||||
'openrouter:openai/o1:prompt': 1500,
|
||||
@@ -515,7 +553,7 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:cohere/command-r7b-12-2024:prompt': 4,
|
||||
'openrouter:cohere/command-r7b-12-2024:completion': 15,
|
||||
'openrouter:meta-llama/llama-3.3-70b-instruct:prompt': 13,
|
||||
'openrouter:meta-llama/llama-3.3-70b-instruct:completion': 39,
|
||||
'openrouter:meta-llama/llama-3.3-70b-instruct:completion': 38,
|
||||
'openrouter:amazon/nova-lite-v1:prompt': 6,
|
||||
'openrouter:amazon/nova-lite-v1:completion': 24,
|
||||
'openrouter:amazon/nova-lite-v1:image': 9000,
|
||||
@@ -541,58 +579,50 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:raifle/sorcererlm-8x22b:completion': 450,
|
||||
'openrouter:thedrummer/unslopnemo-12b:prompt': 40,
|
||||
'openrouter:thedrummer/unslopnemo-12b:completion': 40,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:prompt': 80,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:completion': 400,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:input_cache_read': 8,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:input_cache_write': 100,
|
||||
'openrouter:anthropic/claude-3.5-haiku:prompt': 80,
|
||||
'openrouter:anthropic/claude-3.5-haiku:completion': 400,
|
||||
'openrouter:anthropic/claude-3.5-haiku:web_search': 1000000,
|
||||
'openrouter:anthropic/claude-3.5-haiku:input_cache_read': 8,
|
||||
'openrouter:anthropic/claude-3.5-haiku:input_cache_write': 100,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:prompt': 80,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:completion': 400,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:input_cache_read': 8,
|
||||
'openrouter:anthropic/claude-3.5-haiku-20241022:input_cache_write': 100,
|
||||
'openrouter:anthracite-org/magnum-v4-72b:prompt': 250,
|
||||
'openrouter:anthracite-org/magnum-v4-72b:prompt': 300,
|
||||
'openrouter:anthracite-org/magnum-v4-72b:completion': 500,
|
||||
'openrouter:anthropic/claude-3.5-sonnet:prompt': 300,
|
||||
'openrouter:anthropic/claude-3.5-sonnet:completion': 1500,
|
||||
'openrouter:anthropic/claude-3.5-sonnet:image': 480000,
|
||||
'openrouter:anthropic/claude-3.5-sonnet:input_cache_read': 30,
|
||||
'openrouter:anthropic/claude-3.5-sonnet:input_cache_write': 375,
|
||||
'openrouter:mistralai/ministral-3b:prompt': 4,
|
||||
'openrouter:mistralai/ministral-3b:completion': 4,
|
||||
'openrouter:mistralai/ministral-8b:prompt': 10,
|
||||
'openrouter:mistralai/ministral-8b:completion': 10,
|
||||
'openrouter:mistralai/ministral-3b:prompt': 4,
|
||||
'openrouter:mistralai/ministral-3b:completion': 4,
|
||||
'openrouter:qwen/qwen-2.5-7b-instruct:prompt': 4,
|
||||
'openrouter:qwen/qwen-2.5-7b-instruct:completion': 10,
|
||||
'openrouter:nvidia/llama-3.1-nemotron-70b-instruct:prompt': 60,
|
||||
'openrouter:nvidia/llama-3.1-nemotron-70b-instruct:completion': 60,
|
||||
'openrouter:inflection/inflection-3-productivity:prompt': 250,
|
||||
'openrouter:inflection/inflection-3-productivity:completion': 1000,
|
||||
'openrouter:nvidia/llama-3.1-nemotron-70b-instruct:prompt': 120,
|
||||
'openrouter:nvidia/llama-3.1-nemotron-70b-instruct:completion': 120,
|
||||
'openrouter:inflection/inflection-3-pi:prompt': 250,
|
||||
'openrouter:inflection/inflection-3-pi:completion': 1000,
|
||||
'openrouter:inflection/inflection-3-productivity:prompt': 250,
|
||||
'openrouter:inflection/inflection-3-productivity:completion': 1000,
|
||||
'openrouter:thedrummer/rocinante-12b:prompt': 17,
|
||||
'openrouter:thedrummer/rocinante-12b:completion': 43,
|
||||
'openrouter:anthracite-org/magnum-v2-72b:prompt': 300,
|
||||
'openrouter:anthracite-org/magnum-v2-72b:completion': 300,
|
||||
'openrouter:meta-llama/llama-3.2-3b-instruct:prompt': 2,
|
||||
'openrouter:meta-llama/llama-3.2-3b-instruct:completion': 2,
|
||||
'openrouter:meta-llama/llama-3.2-1b-instruct:prompt': 1,
|
||||
'openrouter:meta-llama/llama-3.2-1b-instruct:completion': 1,
|
||||
'openrouter:meta-llama/llama-3.2-11b-vision-instruct:prompt': 5,
|
||||
'openrouter:meta-llama/llama-3.2-11b-vision-instruct:completion': 5,
|
||||
'openrouter:meta-llama/llama-3.2-11b-vision-instruct:image': 7948,
|
||||
'openrouter:meta-llama/llama-3.2-1b-instruct:prompt': 3,
|
||||
'openrouter:meta-llama/llama-3.2-1b-instruct:completion': 20,
|
||||
'openrouter:meta-llama/llama-3.2-90b-vision-instruct:prompt': 35,
|
||||
'openrouter:meta-llama/llama-3.2-90b-vision-instruct:completion': 40,
|
||||
'openrouter:meta-llama/llama-3.2-90b-vision-instruct:image': 50580,
|
||||
'openrouter:meta-llama/llama-3.2-11b-vision-instruct:prompt': 5,
|
||||
'openrouter:meta-llama/llama-3.2-11b-vision-instruct:completion': 5,
|
||||
'openrouter:meta-llama/llama-3.2-11b-vision-instruct:image': 7948,
|
||||
'openrouter:qwen/qwen-2.5-72b-instruct:prompt': 7,
|
||||
'openrouter:qwen/qwen-2.5-72b-instruct:completion': 26,
|
||||
'openrouter:neversleep/llama-3.1-lumimaid-8b:prompt': 9,
|
||||
'openrouter:neversleep/llama-3.1-lumimaid-8b:completion': 60,
|
||||
'openrouter:openai/o1-mini-2024-09-12:prompt': 110,
|
||||
'openrouter:openai/o1-mini-2024-09-12:completion': 440,
|
||||
'openrouter:openai/o1-mini-2024-09-12:input_cache_read': 55,
|
||||
'openrouter:openai/o1-mini:prompt': 110,
|
||||
'openrouter:openai/o1-mini:completion': 440,
|
||||
'openrouter:openai/o1-mini:input_cache_read': 55,
|
||||
'openrouter:mistralai/pixtral-12b:prompt': 10,
|
||||
'openrouter:mistralai/pixtral-12b:completion': 10,
|
||||
'openrouter:mistralai/pixtral-12b:image': 14450,
|
||||
@@ -624,10 +654,10 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:meta-llama/llama-3.1-405b:completion': 400,
|
||||
'openrouter:meta-llama/llama-3.1-8b-instruct:prompt': 2,
|
||||
'openrouter:meta-llama/llama-3.1-8b-instruct:completion': 3,
|
||||
'openrouter:meta-llama/llama-3.1-405b-instruct:prompt': 350,
|
||||
'openrouter:meta-llama/llama-3.1-405b-instruct:completion': 350,
|
||||
'openrouter:meta-llama/llama-3.1-70b-instruct:prompt': 40,
|
||||
'openrouter:meta-llama/llama-3.1-70b-instruct:completion': 40,
|
||||
'openrouter:meta-llama/llama-3.1-405b-instruct:prompt': 80,
|
||||
'openrouter:meta-llama/llama-3.1-405b-instruct:completion': 80,
|
||||
'openrouter:mistralai/mistral-nemo:prompt': 2,
|
||||
'openrouter:mistralai/mistral-nemo:completion': 4,
|
||||
'openrouter:openai/gpt-4o-mini-2024-07-18:prompt': 15,
|
||||
@@ -640,25 +670,22 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:openai/gpt-4o-mini:input_cache_read': 7,
|
||||
'openrouter:google/gemma-2-27b-it:prompt': 65,
|
||||
'openrouter:google/gemma-2-27b-it:completion': 65,
|
||||
'openrouter:google/gemma-2-9b-it:prompt': 1,
|
||||
'openrouter:google/gemma-2-9b-it:completion': 3,
|
||||
'openrouter:anthropic/claude-3.5-sonnet-20240620:prompt': 300,
|
||||
'openrouter:anthropic/claude-3.5-sonnet-20240620:completion': 1500,
|
||||
'openrouter:anthropic/claude-3.5-sonnet-20240620:image': 480000,
|
||||
'openrouter:anthropic/claude-3.5-sonnet-20240620:input_cache_read': 30,
|
||||
'openrouter:anthropic/claude-3.5-sonnet-20240620:input_cache_write': 375,
|
||||
'openrouter:google/gemma-2-9b-it:prompt': 3,
|
||||
'openrouter:google/gemma-2-9b-it:completion': 9,
|
||||
'openrouter:sao10k/l3-euryale-70b:prompt': 148,
|
||||
'openrouter:sao10k/l3-euryale-70b:completion': 148,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.3:prompt': 3,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.3:completion': 5,
|
||||
'openrouter:mistralai/mistral-7b-instruct:prompt': 3,
|
||||
'openrouter:mistralai/mistral-7b-instruct:completion': 5,
|
||||
'openrouter:nousresearch/hermes-2-pro-llama-3-8b:prompt': 3,
|
||||
'openrouter:nousresearch/hermes-2-pro-llama-3-8b:completion': 8,
|
||||
'openrouter:mistralai/mistral-7b-instruct:prompt': 3,
|
||||
'openrouter:mistralai/mistral-7b-instruct:completion': 5,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.3:prompt': 20,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.3:completion': 20,
|
||||
'openrouter:microsoft/phi-3-mini-128k-instruct:prompt': 10,
|
||||
'openrouter:microsoft/phi-3-mini-128k-instruct:completion': 10,
|
||||
'openrouter:microsoft/phi-3-medium-128k-instruct:prompt': 100,
|
||||
'openrouter:microsoft/phi-3-medium-128k-instruct:completion': 100,
|
||||
'openrouter:meta-llama/llama-guard-2-8b:prompt': 20,
|
||||
'openrouter:meta-llama/llama-guard-2-8b:completion': 20,
|
||||
'openrouter:openai/gpt-4o-2024-05-13:prompt': 500,
|
||||
'openrouter:openai/gpt-4o-2024-05-13:completion': 1500,
|
||||
'openrouter:openai/gpt-4o-2024-05-13:image': 722500,
|
||||
@@ -669,14 +696,12 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:openai/gpt-4o:extended:prompt': 600,
|
||||
'openrouter:openai/gpt-4o:extended:completion': 1800,
|
||||
'openrouter:openai/gpt-4o:extended:image': 722500,
|
||||
'openrouter:meta-llama/llama-guard-2-8b:prompt': 20,
|
||||
'openrouter:meta-llama/llama-guard-2-8b:completion': 20,
|
||||
'openrouter:meta-llama/llama-3-8b-instruct:prompt': 3,
|
||||
'openrouter:meta-llama/llama-3-8b-instruct:completion': 6,
|
||||
'openrouter:meta-llama/llama-3-70b-instruct:prompt': 30,
|
||||
'openrouter:meta-llama/llama-3-70b-instruct:completion': 40,
|
||||
'openrouter:mistralai/mixtral-8x22b-instruct:prompt': 90,
|
||||
'openrouter:mistralai/mixtral-8x22b-instruct:completion': 90,
|
||||
'openrouter:meta-llama/llama-3-8b-instruct:prompt': 3,
|
||||
'openrouter:meta-llama/llama-3-8b-instruct:completion': 6,
|
||||
'openrouter:mistralai/mixtral-8x22b-instruct:prompt': 200,
|
||||
'openrouter:mistralai/mixtral-8x22b-instruct:completion': 600,
|
||||
'openrouter:microsoft/wizardlm-2-8x22b:prompt': 48,
|
||||
'openrouter:microsoft/wizardlm-2-8x22b:completion': 48,
|
||||
'openrouter:openai/gpt-4-turbo:prompt': 1000,
|
||||
@@ -694,10 +719,10 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:anthropic/claude-3-opus:input_cache_write': 1875,
|
||||
'openrouter:mistralai/mistral-large:prompt': 200,
|
||||
'openrouter:mistralai/mistral-large:completion': 600,
|
||||
'openrouter:openai/gpt-4-turbo-preview:prompt': 1000,
|
||||
'openrouter:openai/gpt-4-turbo-preview:completion': 3000,
|
||||
'openrouter:openai/gpt-3.5-turbo-0613:prompt': 100,
|
||||
'openrouter:openai/gpt-3.5-turbo-0613:completion': 200,
|
||||
'openrouter:openai/gpt-4-turbo-preview:prompt': 1000,
|
||||
'openrouter:openai/gpt-4-turbo-preview:completion': 3000,
|
||||
'openrouter:mistralai/mistral-small:prompt': 20,
|
||||
'openrouter:mistralai/mistral-small:completion': 60,
|
||||
'openrouter:mistralai/mistral-tiny:prompt': 25,
|
||||
@@ -708,28 +733,28 @@ export const OPENROUTER_COST_MAP = {
|
||||
'openrouter:mistralai/mixtral-8x7b-instruct:completion': 54,
|
||||
'openrouter:neversleep/noromaid-20b:prompt': 100,
|
||||
'openrouter:neversleep/noromaid-20b:completion': 175,
|
||||
'openrouter:alpindale/goliath-120b:prompt': 400,
|
||||
'openrouter:alpindale/goliath-120b:completion': 550,
|
||||
'openrouter:alpindale/goliath-120b:prompt': 600,
|
||||
'openrouter:alpindale/goliath-120b:completion': 800,
|
||||
'openrouter:openrouter/auto:prompt': -100000000,
|
||||
'openrouter:openrouter/auto:completion': -100000000,
|
||||
'openrouter:openai/gpt-4-1106-preview:prompt': 1000,
|
||||
'openrouter:openai/gpt-4-1106-preview:completion': 3000,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.1:prompt': 11,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.1:completion': 19,
|
||||
'openrouter:openai/gpt-3.5-turbo-instruct:prompt': 150,
|
||||
'openrouter:openai/gpt-3.5-turbo-instruct:completion': 200,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.1:prompt': 11,
|
||||
'openrouter:mistralai/mistral-7b-instruct-v0.1:completion': 19,
|
||||
'openrouter:openai/gpt-3.5-turbo-16k:prompt': 300,
|
||||
'openrouter:openai/gpt-3.5-turbo-16k:completion': 400,
|
||||
'openrouter:mancer/weaver:prompt': 113,
|
||||
'openrouter:mancer/weaver:completion': 113,
|
||||
'openrouter:undi95/remm-slerp-l2-13b:prompt': 45,
|
||||
'openrouter:undi95/remm-slerp-l2-13b:completion': 65,
|
||||
'openrouter:gryphe/mythomax-l2-13b:prompt': 5,
|
||||
'openrouter:gryphe/mythomax-l2-13b:completion': 9,
|
||||
'openrouter:gryphe/mythomax-l2-13b:prompt': 6,
|
||||
'openrouter:gryphe/mythomax-l2-13b:completion': 6,
|
||||
'openrouter:openai/gpt-4-0314:prompt': 3000,
|
||||
'openrouter:openai/gpt-4-0314:completion': 6000,
|
||||
'openrouter:openai/gpt-4:prompt': 3000,
|
||||
'openrouter:openai/gpt-4:completion': 6000,
|
||||
'openrouter:openai/gpt-3.5-turbo:prompt': 50,
|
||||
'openrouter:openai/gpt-3.5-turbo:completion': 150,
|
||||
'openrouter:openai/gpt-4-0314:prompt': 3000,
|
||||
'openrouter:openai/gpt-4-0314:completion': 6000,
|
||||
};
|
||||
|
||||
Reference in New Issue
Block a user