Multimodal analysis
Combine text instructions with images, video, PDFs, or other visual context when the task depends on both perception and reasoning.
Independent tool overview
Muse Spark at a glance
Muse Spark was Meta Superintelligence Labs' first natively multimodal reasoning model, released in April 2026 with tool use, visual reasoning, and multi-agent orchestration. The original model has since been succeeded by Muse Spark 1.1, so current users should evaluate the 1.1 model in Meta AI or the Meta Model API rather than treating the launch version as Meta's latest offering.
Visit the official Muse Spark site ↗
- Developer
- Meta Superintelligence Labs
- Original release
- April 8, 2026
- Current successor
- Muse Spark 1.1
- Current access
- Meta AI and Meta Model API public preview
- Core format
- Multimodal reasoning and agentic model
Overview
What Muse Spark is
Meta introduced Muse Spark in April 2026 as the first model in its new Muse family. It was built to reason across text and visual information, use tools, and coordinate multiple agents rather than operate as a text-only chatbot.
The original release powered Meta AI and introduced Contemplating mode, which scales difficult questions across parallel reasoning agents. Meta positioned it for visual troubleshooting, interactive artifacts, wellness education, and other tasks where perception and reasoning need to work together.
Muse Spark is no longer the newest model in the family. Meta released Muse Spark 1.1 in July 2026 as a significant upgrade and says it now powers Meta AI. Version 1.1 adds stronger computer use, coding, tool use, a one-million-token context window, and public-preview developer access through the Meta Model API.
This page remains active because Muse Spark is an ongoing model family, not an abandoned product. However, benchmark claims from the April launch describe the original model and should not be assumed to represent the current 1.1 deployment or every product surface.
Use cases
Who Muse Spark is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Long-running agent workflows
Use Muse Spark 1.1 for tool-rich projects that require planning, delegation, context management, and adaptation across multiple steps.
Coding and computer use
Evaluate the 1.1 successor for codebase work, browser or desktop workflows, visual debugging, and structured tool execution.
Capabilities
Core Muse Spark features
Native multimodal reasoning
The model was designed to integrate visual information with language, reasoning, and tools rather than bolt image understanding onto a text-only workflow.
Tool use
Muse Spark can call external capabilities as part of a larger plan; Muse Spark 1.1 extends this with developer-facing tool and function calling.
Multi-agent orchestration
Contemplating mode and the 1.1 agent architecture can divide difficult work among parallel agents and consolidate the result.
Visual artifact generation
Meta demonstrates the model turning visual context into interactive explanations, annotations, minigames, and web interfaces.
Computer use in 1.1
The successor can choose between interface actions and generated scripts while operating across changing, multi-application workflows.
One-million-token context in 1.1
Meta says the current model can manage a one-million-token context, retrieve earlier information, and compact long sessions while preserving critical steps.
Process
How the Muse Spark workflow works
Step 1
Use the current model
Open Meta AI or the Meta Model API and confirm that the selected surface is using Muse Spark 1.1, not the original April launch model.
Step 2
Provide the relevant context
Attach only the images, video, documents, or tool permissions needed for the task and explain the desired output and constraints.
Step 3
Bound agent permissions
For API or computer-use deployments, use narrow tool allowlists, isolated workspaces, approval gates, spending limits, and reversible actions.
Step 4
Verify the result
Review citations, calculations, code, visual localization, and external actions before using the output in health, security, financial, legal, or production decisions.
Cost
Muse Spark pricing and free plan
Meta's cited launch materials do not publish a standalone consumer price for Muse Spark or a public per-token rate for the Meta Model API preview. Access terms, availability, limits, and API billing can vary by account and product surface, so users should confirm the current terms shown in Meta AI or the developer console before committing a workload.
Meta AI
No standalone Muse rate disclosed
Consumer access to the current Muse Spark 1.1 model in Thinking mode.
- Available through Meta AI and the Meta AI app
- Product availability and usage limits can vary
- Contemplating or advanced modes may roll out gradually
Meta Model API
Public-preview terms
Developer access to Muse Spark 1.1 through Meta's new model API.
- Public preview announced in July 2026
- Tool and function calling are supported
- Confirm current usage pricing and quotas in the developer console
Pricing checked . Check current pricing at the source ↗
Assessment
Muse Spark strengths and limitations
Where it stands out
- Combines multimodal perception, reasoning, tool use, and agent orchestration in one model family
- Muse Spark 1.1 adds stronger coding, computer use, long-context management, and developer access
- Parallel-agent reasoning can spend more compute on difficult tasks without relying only on one long serial chain
- Meta publishes detailed evaluation and safety reports for both the original and 1.1 releases
- Available in a consumer assistant as well as a developer API preview
What to consider
- The original Muse Spark was superseded by Muse Spark 1.1, so launch benchmarks and capability descriptions are not the current product specification
- Meta explicitly identified long-horizon agentic systems and coding workflows as gaps in the original release
- Agent and computer-use systems can take consequential actions, encounter prompt injection, or propagate an early mistake across later steps
- Model-generated health explanations, scores, code, citations, and visual annotations still require qualified human verification
- Meta's public launch pages do not provide a complete price and quota table for every Meta AI and API surface
- Availability and advanced modes may vary by account, geography, rollout stage, and connected tools
Compare
Muse Spark alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Business Operations
ChatGPT
Consider ChatGPT for another multimodal assistant with reasoning, coding, research, and connected-tool workflows.
Explore ChatGPT →Content Creator
Gemini
Consider Gemini when Google Workspace integration and Google's multimodal model ecosystem are a stronger fit.
Explore Gemini →Project Management
Claude
Consider Claude for long-context document work, coding, and agentic development workflows in Anthropic's ecosystem.
Explore Claude →Questions
Muse Spark FAQs
Is Muse Spark still active?
Yes, as a model family. The original April 2026 model has been succeeded by Muse Spark 1.1, which Meta says now powers Meta AI and is available to developers through the Meta Model API public preview.
What is the difference between Muse Spark and Muse Spark 1.1?
Muse Spark introduced Meta's natively multimodal reasoning and multi-agent direction. Version 1.1 is the current upgrade, with stronger tool and computer use, coding, multimodal understanding, one-million-token context management, and developer API access.
Can developers use Muse Spark?
Developers can use Muse Spark 1.1 through the Meta Model API, which Meta launched in public preview with tool and function calling and developer-controlled prompts.
What is Contemplating mode?
It is Meta's multi-agent reasoning mode for difficult tasks. It coordinates parallel agents so the system can spend more test-time compute and combine several lines of reasoning.
Is Muse Spark safe to run autonomously?
No agentic model should receive broad, unsupervised authority by default. Meta's own 1.1 report recommends application-level controls such as strict tool allowlists and workspace isolation; high-impact actions should also require human approval.
Bottom line
Our Muse Spark verdict
Muse Spark is best understood as the foundation of Meta's current Muse model family, not as a frozen one-off tool. The April release established the multimodal and multi-agent direction, while Muse Spark 1.1 is the version to evaluate today for real use. Its strongest fit is a carefully controlled workflow where visual context, long memory, coding, and tools need to work together; its outputs and actions still need human oversight.
Visit Muse Spark website ↗Similar Consumer Tools
View all Consumer AI tools →
GLM-5.1
GLM-5.1 — Z AI's new open-source flagship model built for long-horizon agentic coding
HeyClicky (formerly Clicky)
Clicky - Open-source AI teacher that lives next to your cursor
Gemma 4
Gemma 4 - Google's new open-weight reasoning and agentic model family across four sizes
.png)
Gemini for Mac
Gemini for Mac - Google's new native macOS app for Gemini
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.