Muse Spark
Meta Superintelligence Labs' first model — a natively multimodal reasoning model with contemplating mode and multi-agent orchestration
Muse Spark
Meta • April 2026
Training Data
Not disclosed
Muse Spark
April 2026
Parameters
Not disclosed
Training Method
RL with test-time compute, rebuilt pretraining stack
Context Window
Not disclosed
Knowledge Cutoff
Not disclosed
Key Features
Contemplating Mode • Visual Chain of Thought • Multi-Agent Orchestration • Native Multimodal
Capabilities
Reasoning: Excellent (HLE 58%)
Multimodal: Outstanding
Health: Excellent (physician-trained)
Efficiency: ~10× vs Llama 4 Maverick
What's New in This Version
First MSL model under Alexandr Wang — native multimodal reasoning with test-time compute and order-of-magnitude efficiency gain over Llama 4 Maverick
Meta Superintelligence Labs' first model — a natively multimodal reasoning model with contemplating mode and multi-agent orchestration
What's New in This Version
First MSL model under Alexandr Wang — native multimodal reasoning with test-time compute and order-of-magnitude efficiency gain over Llama 4 Maverick
Technical Specifications
Key Features
Capabilities
Other Meta Models
Explore more models from Meta
Muse Spark Open
Meta's open-weight frontier model — an 816B mixture-of-experts with 42B active, natively multimodal, built for coding and long-horizon agents, and able to run on two Mac Studios
Muse Wonder
Meta's first frontier-class model, closed-weight and built for personal and professional agents, with Active Memory across conversations and Active Focus on long goals; its larger sibling, Muse Wonder Full, is reserved for paid Muse Code subscribers
Muse Spark 1.3
Meta's most capable Muse model, built for long-horizon agentic work — it collaborates, asks before consequential actions, and holds complex instructions across multi-step tasks
