Robots Atlas>ROBOTS ATLAS
Muse Spark

Muse Spark

Family: Muse
Natively multimodal reasoning model by Meta Superintelligence Labs — first in the Muse family, with tool-use, visual chain of thought, and multi-agent orchestration.
✓ Active⏳ Limited accessLLMMultimodalReasoning modelTool-using model📁 Muse
Release date
8 April 2026
Access:APIHostedDeployment:☁ Cloud

Overview

Muse Spark is the first model in the new Muse series developed by Meta Superintelligence Labs. Meta describes it as a small, fast model designed for reasoning tasks and multimodal understanding, currently used in Meta AI and being gradually deployed across the company's products.

Classification
LLMMultimodalReasoning modelTool-using model
Family: Muse
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
Tools
📥 Input: text, image

Technical specification

License
Proprietary
Features:Tool use
Modalities
⬇ Input
textimage
⬆ Output
text

Capabilities and applications

Native model capabilities
Reasoning
The model's ability to reason logically and solve complex problems.
Category: reasoning
Multimodal understanding
Category: multimodal
Image understanding
Analysing and interpreting the content of images.
Category: vision
Chart understanding
Reading and interpreting charts, tables and diagrams.
Category: vision
Coding
Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.
Category: coding

Benchmark results

2 benchmarks
Humanity's Last Exam
accuracy · Contemplating mode (multi-agent parallel reasoning)
58%
📅 8 Apr 2026📄 Meta AI Blog — Introducing Muse Spark: Scaling Towards Personal Superintelligence
Contemplating mode orchestrates multiple agents reasoning in parallel. Competes with Gemini Deep Think and GPT Pro extreme reasoning modes.
FrontierScience Research
accuracy · Contemplating mode (multi-agent parallel reasoning)
38%
📅 8 Apr 2026📄 Meta AI Blog — Introducing Muse Spark: Scaling Towards Personal Superintelligence
Contemplating mode result. Full methodology in Meta's methodology document linked from the blog post.

Technical architecture

Core Architecture
Training Techniques