Muse Ember
Neural Network
Muse Ember by Meta is a small, cost-effective model for reasoning, function calling, and repetitive agent tasks.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 1 048 576 tokens
- Max answer length
- 943 718 tokens
- Reasoning effort
- minimal, low, medium, high, extra high, max
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
import OpenAI from 'openai';
const openai = new OpenAI({
apiKey: '<your bothub access token>',
baseURL: 'https://openai.bothub.chat/v1'
});
// Sync - Text generation
async function main() {
const chatCompletion = await openai.chat.completions.create({
messages: [{ role: 'user', content: 'Say this is a test' }],
model: 'muse-ember',
});
}
// Async - Text generation
async function main() {
const stream = await openai.chat.completions.create({
messages: [{ role: 'user', content: 'Say this is a test' }],
model: 'muse-ember',
stream: true
});
for await (const chunk of stream) {
const part: string | null = chunk.choices[0].delta?.content ?? null;
}
}
main();
How it works Muse Ember?
Muse Ember is Meta's junior and most cost-effective model, designed for complex reasoning and daily agent routines. Tool calling, specific coding agent subtasks, and request routing between models are where Ember excels, as these calls are repeated thousands of times and require predictable costs. Via BotHub, the model is accessible from Russia without a VPN or foreign card: payment is in rubles, pay-as-you-go for tokens rather than a subscription, and purchased tokens do not expire. You can switch to another model or compare responses in the same window, and a unified OpenAI-compatible API allows integrating Ember into a multi-model system without rewriting the integration. Data is protected by AES-GCM encryption and is not saved; for teams, contracts, invoices, EDI, and limits in the admin panel are available. Typical scenarios: orchestrating agent pipelines that distribute tasks to larger models, code generation and refactoring with tools, log analysis and step-by-step debugging, request routing within products, and maintaining internal bots where cost per call is critical.Frequently asked questions about Muse Ember
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
muse-ember is a text model: it writes and edits text, parses input documents and images, reasons before answering, and calls functions. It is suitable for working with code, long documents, analytics, and agent scenarios where you need to maintain a large context volume and get a structured result.
According to the card data, the maximum volume of a single response is 943,718 tokens, and the input context reaches 1,048,576 tokens. The actual length depends on your request and settings on the BotHub side: it is more convenient to collect long responses in parts, especially if you need to parse a large document.
Yes, muse-ember has a reasoning mode before answering. The model can first think through the solution steps and then provide the result, which helps with tasks involving logic, calculations, and multi-step instructions. Check the BotHub interface to see if this mode is enabled via a separate switch.
Yes, both function calling and structured JSON output are supported. This means the model can be connected to external services and receive responses in a format that code can parse immediately, without manual editing.
You can input images and documents — the model will parse the content and answer based on it. Audio and video input is not noted in our data, so it is better to convert such files into text or separate frames in advance.
There is no separate quality assessment for Russian in the model data, so we will not promise a specific level. It is easiest to check with your own tasks: send a typical request and compare the response with other models in BotHub — switching takes a couple of clicks.