Muse Spark Open
Neural Network
Muse Spark Open is an open-weights model for agentic programming, tool calling, and codebase work.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 1 048 576 tokens
- Max answer length
- 943 718 tokens
- Reasoning effort
- low, medium, high, extra high
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Muse Spark Open?
Muse Spark Open is an open-weights model trained for development tasks: agentic coding scenarios, reliable tool calling, and understanding large codebases as a whole rather than fragments. Its strength is high first-attempt accuracy, where fixes must work immediately without extra iterations. Through BotHub, you connect without VPN or foreign cards: pay with Russian cards in rubles based on usage; tokens do not expire. Over 250 models are available in one window, making it easy to switch and compare answers, while a unified OpenAI-compatible API allows switching models in multi-model systems without rewriting integrations. For teams, we offer contracts, invoices, electronic document management, an admin panel with limits, and data encrypted via AES-GCM without storage. Use Muse Spark Open for agents that edit repositories and run tests, for refactoring and analyzing unfamiliar projects, for debugging and finding bug causes, for generating code with external tool calls, and for reviewing changes before merging.Frequently asked questions about Muse Spark Open
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.
This is a text model from Meta: it answers queries, parses documents and images, reasons before answering, and calls external functions. It is suitable for analyzing large materials, working with code, research tasks, and agentic scenarios requiring long context and tools.
The output limit is 943,718 tokens per response, and the context window is 1,048,576 tokens. This is enough for large documents, long code dumps, or detailed reports without splitting the task into dozens of separate requests.
Yes, reasoning mode is supported: the model breaks down the task step-by-step before providing a result. This significantly helps with logic, mathematics, code analysis, and questions where it is important to arrive at the answer sequentially rather than guessing.
Yes, the model can call functions and return structured JSON. Through the unified OpenAI-compatible BotHub API, you can connect it to your services and agents, and switch to another model if needed without rewriting the integration.
According to our data, text, images, and documents are accepted as input, including native formats. The model will read a screenshot, diagram, PDF, or contract and parse the content. There is no information in our data regarding audio and video.
Our data contains no information about languages, so we cannot evaluate this in advance — it is more reliable to test it on your own tasks. BotHub has a Russian interface, accepts ruble card payments, and provides access from Russia without VPN or foreign cards.