Meta Superintelligence Labs · Open weights
Muse Spark OpenOpen-weight AI for coding and agents
An open-weight multimodal model from Meta Superintelligence Labs. 816B total parameters, about 42B active per token and a 1M-token context window for coding, agents and extended tasks.
- Token context window
- 1M
- Total model parameters
- 816B
- Active parameters per token
- 42B
Review this production incident, identify the likely root cause, and propose the safest recovery plan.
Prompt sent to Muse Spark Open
Muse Spark Open
Open-weight multimodal model
// Illustrative workflow, not a performance measurement. Published model specifications and the playground’s current limits are different; see the guide before testing.
Muse Spark Open playground
Muse Spark Open playground
Try Muse Spark Open in the playground with text and supported image or video URLs. Compare reasoning settings; features depend on the connected model.
Account credits required
Muse Spark Open
Our configured model
What would you like to explore?
Ask an open-ended question, review code, reason across a long document, or attach a public image or video URL.
Try a field test
Choose your path
Explore. Deploy. Build.
Start with the model card, choose a hosted or self-managed runtime, then evaluate the model on your own coding and agent tasks.
- 01
Understand the model
Review the 816B / 42B architecture, 1M context window, input modalities and Muse Model License before choosing a deployment.
❯ total: 816B / active: 42B / context: 1M - 02
Choose an inference framework
Compare SGLang, vLLM, TensorRT-LLM, Transformers, Apple MLX, llama.cpp and Unsloth against your hardware and workload.
❯ runtime = "your selected framework" - 03
Evaluate a real workflow
Test repository changes, multi-step tool use and long-context tasks. Measure quality, memory use and latency on your own deployment.
❯ evaluate(coding, tools, long_context)
Published model capabilities
What Muse Spark Open can do
Muse Spark Open brings open weights, multimodal understanding and extended reasoning to developers, infrastructure teams and researchers.
deployment = { runtime: selected }
An open inference ecosystem
Run with SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp or Unsloth. Match versions and memory requirements to your deployment.
42B
Active parameters per token
816B total, 42B active
A mixture-of-experts model with about 42B parameters active per token and a 1M-token context window.
- text
- image
- video
- audio
Four native input modalities
The model accepts text, images, video and audio, and produces text. Available inputs depend on the interface and runtime you use.
tools = [get_file, search_records] response = chat({ model, messages, tools }) await execute_approved(response.tool_calls)
Coding and sequential tool use
Build agents that plan, inspect code, call tools and recover from errors across extended tasks. Your application controls execution.
$ curl -X POST /api/v1/chat/completions -H "Authorization: Bearer $MUSE_SPARK_API_KEY" Chat Completions example
Our operated API service
Connect your application to POST /api/v1/chat/completions with an API key. Use standard messages, reasoning settings and streaming responses.
Model downloads and developer resources
Find official weights, the model card, supported inference frameworks and practical notes for deployment and evaluation.
Put it to work
A long-context model for more than chat
Explore Muse Spark Open for repository reasoning, multimodal analysis and long-horizon agents. These snippets illustrate a compatible deployment; adapt the model ID and request format to your runtime.
response = client.chat.completions.create(
model=DEPLOYED_MODEL_ID,
messages=[{"role": "user", "content": repo_context}],
reasoning_effort="high",
)
review = response.choices[0].message.content// trace
- context_window1,000,000
- reasoning_efforthigh
Designed for developers
Explore Muse Spark Open API access
Integrate Muse Spark Open through our OpenAI-compatible Chat Completions API. Input: $1.25 per 1M tokens. Output: $4.25 per 1M tokens (USD). Use your API key with familiar message, reasoning and streaming parameters.
- POST /api/v1/chat/completions
- OpenAI-compatible requests and responses
- Bearer API key authentication
const response = await fetch(
"https://themusesparkopen.com/api/v1/chat/completions", {
method: "POST",
headers: {
"Authorization": `Bearer ${process.env.MUSE_SPARK_API_KEY}`,
"Content-Type": "application/json"
},
body: JSON.stringify({
model: "meta/muse-spark-open",
messages: [{
"role": "user",
"content": "Review this design"
}],
max_tokens: 2000,
reasoning: { effort: "medium" }
})
})
const result = await response.json();
console.log(result.choices[0].message.content);Chat Completions example — Our endpoint · Your API key
# Deployment inputs — configure your chosen runtime
# 1. official weights
export MODEL_REPO="meta-models/Muse-Spark-Open-816B-A42B"
# 2. choose a supported runtime
export INFERENCE_ENGINE="sglang"
Your application keeps control
The model can request a tool call, but your code still validates permissions, arguments, side effects and human approval boundaries.
Agent integration
From open weights to an agent runtime
Deploy Muse Spark Open with a supported framework, then connect tools and evaluate extended tasks in your own application.
- 1
Find official model weights
Use Meta’s Hugging Face collection and review the license and model card.
- 2
Match the runtime to your hardware
Check SGLang, vLLM, TensorRT-LLM, Transformers, MLX, llama.cpp or Unsloth for the required versions and configuration.
- 3
Offer only approved tools
Define the functions your agent may request, then validate arguments and permissions before execution.
Documentation
Muse Spark Open downloads and deployment
A starting point for AI developers, agent builders, inference teams and model evaluators.
Model overview
Understand the open-weight release, 816B / 42B parameters and 1M context.
Multimodal inputs
Review native text, image, video and audio support alongside interface-specific limits.
Run and deploy
Compare the seven announced frameworks and follow official setup resources.
Download and license
Find official weights and read the Muse Model License before deployment.
Plans and usage
Muse Spark Open Pricing & Credits
Compare Muse Spark Open credit plans for model evaluation, coding and agent workflows. Chat Completions API requests use your account credits.
Monthly subscriptions include credits valid for three months. One-time credit packs never expire and do not renew automatically.
Muse Spark Open token pricing
- Input
- $1.25USD per 1M tokens
- Output
- $4.25USD per 1M tokens
FAQ
Muse Spark Open FAQs
Model facts and access details for Muse Spark Open.
Open weights. Your next workflow.
Explore Muse Spark Open, test a coding prompt and use the developer guide to plan your deployment or API integration.
Open the playground