Muse Spark 1.1 by Meta AI
Muse Spark is a natively multimodal reasoning model from Meta Superintelligence Labs, the first in the Muse family. It features tool-use, visual chain of thought, and multi-agent orchestration, and is available at meta.ai and the Meta AI app.
Key features
- ✓Natively multimodal reasoning (visual, text)
- ✓Tool-use and visual chain of thought
- ✓Contemplating mode: parallel multi-agent reasoning
- ✓Health reasoning with physician-curated training data
- ✓Efficient pretraining with order of magnitude compute reduction
- ✓Reinforcement learning for improved pass rates
Use cases
- →Create interactive games from images (e.g., sudoku)
- →Troubleshoot home appliances with dynamic annotations
- →Personalized health recommendations based on dietary restrictions
- →Yoga form analysis and side-by-side comparison
- →Visual STEM question answering and entity recognition
Pros
- +Competitive multimodal perception, reasoning, health, and agentic tasks
- +Efficient pretraining requiring order of magnitude less compute than previous model
- +Smooth and predictable RL scaling with log-linear improvements
- +Strong safety evaluations with safe margins across all frontier risk categories
- +Contemplating mode achieves 58% on Humanity's Last Exam
Cons
- −Performance gaps in long-horizon agentic systems and coding workflows
- −Contemplating mode rolling out gradually, not immediately available
- −Private API preview limited to select users
Positioning
- Core value: First step toward personal superintelligence, integrating multimodal perception, reasoning, health, and agentic tasks.
- Ideal for: Developers, researchers, general users
- Product type: Web app, Mobile app, API (private preview)
Frequently asked questions
What is Muse Spark?
Muse Spark is a natively multimodal reasoning model from Meta Superintelligence Labs, the first in the Muse family. It features tool-use, visual chain of thought, and multi-agent orchestration.
How can I access Muse Spark?
Muse Spark is available at meta.ai and the Meta AI app. A private API preview is open to select users.
What is Contemplating mode?
Contemplating mode orchestrates multiple agents that reason in parallel, allowing Muse Spark to compete with extreme reasoning modes of frontier models like Gemini Deep Think and GPT Pro. It is rolling out gradually.
What are Muse Spark's capabilities in health?
Muse Spark's health reasoning was trained with over 1,000 physicians. It can generate interactive displays for nutritional content, exercise muscle activation, and more.
How does Muse Spark compare to Llama 4 Maverick in efficiency?
Muse Spark requires over an order of magnitude less compute to reach the same capabilities as Llama 4 Maverick.
What safety evaluations were conducted?
Meta followed the Advanced AI Scaling Framework, evaluating Muse Spark across frontier risk categories, behavioral alignment, and adversarial robustness. The model falls within safe margins for its deployment context.
You may also like
Curated suggestions based on similar categories.
SmartRolePath
SmartRolePath is an AI career intelligence platform that analyzes a user's skills against live hiring data from 50,000+ job postings to generate personalized career paths with real salary data, market demand scores, automation risk assessments, and month-by-month transition roadmaps.
agents-cli
agents-cli is a CLI tool and skill set for building and deploying agents on Google Cloud, designed to integrate with coding agents for rapid development.
OpenJobs AI
Metix AI (formerly OpenJobs AI) is an AI-native hiring platform that delivers interview-ready candidates within 24 hours. You define the role, and the platform handles sourcing, outreach, and screening, with a delivery team ensuring quality. You only pay for candidates who are qualified and ready to interview.
Dim0
Dim0 is an open-source, real-time collaborative AI canvas that combines notes, mini-apps, and AI agents on an infinite board. It is privacy-first, supports multiple AI models, and can be self-hosted.
ForgeAI
ForgeAI builds and deploys enterprise AI solutions with built-in explainability, auditability, and data ownership. It enables rapid validation of AI ideas in two weeks using synthetic data, retrieves conceptually relevant information with calibrated confidence scores, and allows AI to scale across domains (finance, HR, legal, real estate, travel) without retraining.
Suparagent AI
Suparagent AI is an all-in-one AI workspace that combines chat, deep research, presentation slides, autonomous agents, image generation, video creation, and motion graphics into a single platform, powered by multiple leading AI models like GPT-4o, Claude 3.5, and Gemini 2.0.