Description
Muse Spark 1.3 is an AI model from Meta, built for agentic workflows and tuned for competitive coding performance. It is the latest version in Meta's Muse Spark line, following Muse Spark 1.2 and 1.1.
Read More
The model is trained for long-horizon, agentic tasks: it tracks context and prior results across a session, works through messy or conflicting inputs, and asks for clarification when it needs it rather than guessing. Meta reports higher first-attempt accuracy and more reliable tool calling than the previous version, along with results that compete with other frontier models on public evals covering agentic long-context work, agentic browsing, instruction following, and long-horizon coding.
Muse Spark 1.3 has native multimodal perception, meaning it takes in video, images, and documents, and its visual reasoning runs through an actual execution environment rather than following scripted steps, so it can act directly on a screenshot or a short clip. It has a 1 million token context window. Meta pairs the model with Muse Code, a terminal-based multi-agent coding tool built specifically for Muse Spark, and highlights use cases such as multi-agent orchestration, where a team of agent roles turns a one-line product idea into a working app, computer use, autonomous GitHub agents, and search grounding for live web access.
Access comes two ways: directly through the Meta Model API, priced per million tokens, with a contributor version at $0.10 input, $0.002 cached input, and $0.20 output, whose usage is used to improve Meta's products, and a non-contributor version at $1.25, $0.15, and $4.25 that is not; or through OpenRouter for developers who want to swap it into existing tools.
Muse Spark 1.3 is aimed at developers building coding agents, multi-agent systems, and tools that need strong tool use and long context, rather than at casual chat use. It sits alongside Meta's other models, including Muse Glimmer, Muse Image, and Muse Voice Transcribe, in the broader Muse family available through the Meta Model API.
- CategoryAI Models & LLMs
- Pricing ModelUsage-based
- DeploymentCloud
- PlatformsWeb
Muse Spark 1.3 Highlights
Tuned for long-horizon, agentic workflows with reliable tool calling
Competitive coding performance against other frontier models on public evals
Native multimodal perception of video, images, and documents
Visual reasoning through a real execution environment
1 million token context window
Paired with Muse Code, a terminal-based multi-agent coding tool
Available via the Meta Model API or through OpenRouter
Two pricing tiers depending on whether usage improves Meta's products
Getting Started with Muse Spark 1.3
Get API access: sign up for the Meta Model API or access Muse Spark 1.3 through OpenRouter
Point your client at the API: use an OpenAI SDK-compatible client to send your first request
Choose a model variant: pick the contributor-priced or standard-priced version of muse-spark-1.3
Build agentic workflows: use multi-agent orchestration, computer use, or search grounding as needed
Pair with Muse Code: use Meta's terminal tool for multi-agent coding built for Muse Spark
Muse Spark 1.3's Use Cases
- Coding agents
- Multi-agent product builds
- GitHub automation
- Computer-use tasks
