Muse Spark open - Free Promo

Coding-optimized open model for long-horizon agentic work with 1M context

Try now

Read Docs

About model

Free until [END DATE, TIME] PT. After that, this endpoint automatically routes to the paid endpoint and usage is billed at standard rates. Rate limits apply, and we may change or end the offer.

Muse Spark is the coding-optimized open weights model from Meta, trained specifically for agentic coding workflows: higher first-attempt accuracy, reliable tool calling, and deep codebase understanding across the repositories developers actually work in. Meta describes it as the most capable US open-weights model to date. Trained on real developer workflows, from multi-file refactors to debugging sessions and long-running tasks, it holds entire projects in a 1M-token context window, including complex dependencies, legacy code, and thousands of files. Built from the ground up as an agentic model, it plans, acts, observes, and self-corrects over long horizons, browsing the web with multimodal browsing, spawning and coordinating subagents, and using computer-use capabilities unified across desktop, mobile, and browser. As a reasoning model with audio and audio-video perception alongside image and document understanding, it works across what it sees, hears, and reads. Available on Together AI.

‍

Context Window

Entire projects with complex dependencies, legacy code, and thousands of files

Multimodal Perception

Native

Works across what it sees, hears, and reads: images, documents, audio, and video

Intent Commentary

Built-In

States what it is doing and why before acting on long-horizon tasks

Model key capabilities

Agentic Coding: Trained on real developer workflows, from multi-file refactors to debugging sessions and long-running tasks, for higher first-attempt accuracy and reliable tool calling
Long-Horizon Autonomy: Plans, acts, observes, and self-corrects over long horizons, coordinating subagents for decomposable work and stating intent in separate commentary before acting
Multimodal Perception: Audio and audio-video perception alongside image and document understanding, reasoning across what it sees, hears, and reads
Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure

API usage

cURL
Python
Typescript

Endpoint:

meta-models/muse-spark-open-816b-promo

curl -X POST "https://api.together.xyz/v1/chat/completions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-models/muse-spark-open-816b-promo",
    "messages": [
      {
        "role": "user",
        "content": "What are some fun things to do in New York?"
      }
    ]
}'

from together import Together

client = Together()

response = client.chat.completions.create(
  model="meta-models/muse-spark-open-816b-promo",
  messages=[
    {
      "role": "user",
      "content": "What are some fun things to do in New York?"
    }
  ]
)
print(response.choices[0].message.content)

import Together from 'together-ai';
const together = new Together();

const completion = await together.chat.completions.create({
  model: 'meta-models/muse-spark-open-816b-promo',
  messages: [
    {
      role: 'user',
      content: 'What are some fun things to do in New York?'
     }
  ],
});

console.log(completion.choices[0].message.content);

Model card
Architecture Overview:
• Reasoning model with native multimodal perception spanning images, documents, audio, and audio-video input
• 1M-token context window for whole-project codebase understanding
• Full architecture details arrive with the launch technical materials

Training Methodology:
• Trained specifically for agentic coding workflows on real developer work: multi-file refactors, debugging sessions, and long-running tasks
• Trained to browse the web, including multimodal browsing, and to work within an agentic harness: spawning and coordinating subagents, and using computer-use capabilities unified across desktop, mobile, and browser
• Trained on long-horizon, multi-site tasks, generating separate commentary that states intent before acting

Performance Characteristics:
• Benchmarks and performance data arrive with the launch announcement
‍
Prompting
Together AI API Access:
• Access Muse Spark via Together AI APIs using the endpoint meta-models/muse-spark-open-816b-promo
• Authenticate using your Together AI API key in request headers
• Supports text, image, audio, and video input across a 1M-token context window
• The model emits separate intent commentary on long-horizon tasks, so downstream systems can surface what it is doing and why
• Available on Together AI serverless and dedicated infrastructure
‍
Applications & use cases
Agentic Software Engineering:
• Hand off multi-file refactors, debugging sessions, and long-running engineering tasks end to end
• Hold entire repositories, dependencies, and legacy code in the 1M-token window without splitting work into pieces
• Chain reliable tool calls across long trajectories with self-correction built in

Computer Use & Web Automation:
• Automate workflows across desktop, mobile, and browser with unified computer-use capabilities
• Run long-horizon, multi-site tasks with the model browsing the web, including multimodal browsing
• Coordinate subagents for decomposable work under a single orchestrating session

Multimodal Document & Media Understanding:
• Reason over documents, images, audio, and video alongside code and text in one request
• Ground agent decisions in what the model sees, hears, and reads
• Surface transparent intent commentary for auditable automation
‍