Muse Spark Open inference

Put Meta’s open-weight, 816B-parameter mixture-of-experts model to work on coding and agentic tasks. About 42B parameters are active per token.

Test your own workflows on Nebius Token Factory and compare the quality, latency and cost you observe.

Built for coding work

Try debugging, multi-file refactors and codebase questions.

Designed for agents

Evaluate tool calls and multi-step tasks with your own prompts and tools.

816B total, about 42B active

Explore Meta’s mixture-of-experts model on your workloads.

Open weights from Meta

Evaluate the model with access to its published weights and terms.

Measure the whole workflow

Compare outcomes, retries, latency and cost across complete tasks.

Text, image and video

Use text, images and video as inputs through Token Factory’s API.

Featured model

Muse Spark Open

Meta’s coding-focused, open-weight model for developer and agentic applications.

Supports text, image and video input through Token Factory’s API.

Served in FP4.

Best for: Coding assistants, debugging, codebase analysis and tool-driven workflows.

See how it handles your coding workflow

Give the model tasks your team actually runs: a bug spanning several files, a refactor with dependencies, a codebase question or a tool-driven sequence.

Check the result and tool behavior, then measure latency and billed cost across the complete task.

Try the model before integrating

Use the Token Factory Playground to test prompts and inspect responses. Start with a representative task, then refine your prompts before moving to the API.

Full screen image

Build with a familiar API

Connect through Token Factory’s OpenAI-compatible API. Use the model ID and supported request options shown on the live model detail page.

Run Muse Spark Open on Token Factory

Start with serverless inference to test your workload and scale usage as needed. Check the live model details for availability and pricing.

Frequently asked questions

Meta’s open-weight mixture-of-experts model, positioned for coding and agentic workflows.