AI News (2026/4/9): Meta Launches Native Multimodal Large Model Muse Spark
Executive Summary:
Meta Super Intelligence Lab (MSL) has launched its first native multimodal large model, Muse Spark. The model's score on Artificial Analysis has jumped from 18 points for Llama 4 to 52 points, trailing only GPT-5.4 and Gemini 3.1 Pro. Muse Spark excels in visual understanding and health question-answering, driving Meta's stock price up by nearly 10%.
News Details
Meta Super Intelligence Lab (MSL) launched its first native multimodal large model, Muse Spark, on April 9. The model achieved a score of 52 points on Artificial Analysis, significantly higher than LLaMA 4's 18 points, and only slightly behind GPT-5.4 and Gemini 3.1 Pro. This technological breakthrough not only enhances Meta's competitiveness in the AI field but also drove its stock price up by nearly 10% after the release.
Key Points
Visual Chain of Thought: Muse Spark employs a native multimodal reasoning architecture that can process both image and text data, and perform complex reasoning tasks through the Visual Chain of Thought. This feature enables it to excel in the CharXiv visual understanding test.
Multi-Agent Orchestration: The model supports multi-agent orchestration, allowing multiple AI agents to work together to complete complex tasks. This design increases the model's flexibility and adaptability, making it advantageous in various application scenarios.
Reflection Mode: Muse Spark introduces "Reflection Mode", a deep thinking mechanism that enables the model to perform multiple iterations of reasoning when faced with complex problems, thereby improving accuracy and reliability. This mode is particularly prominent in the HealthBench health question-answering test.
AI-ALL In-Depth Commentary
The launch of Muse Spark marks a significant advancement for Meta in the field of multimodal AI. Its Visual Chain of Thought and multi-agent orchestration capabilities not only enhance the model's reasoning abilities but also provide developers with a richer toolset. Particularly, its performance in health question-answering and visual understanding demonstrates its potential in practical applications. However, compared to GPT-5.4 and Gemini 3.1 Pro, Muse Spark still has room for improvement. In the future, with the involvement of more developers and researchers, Muse Spark is expected to be further optimized and refined, becoming a key player in the AI ecosystem.
