Meta launches Muse Spark: New AI model outperforming GPT, Gemini, and Grok in multimodal reasoning

Meta has launched its first new model in the Muse Family, called Muse Spark, developed by Meta Superintelligence Labs. The model is designed as a multimodal reasoning model, which can produce several texts and images. In addition, it will also use tools to solve complex problems. According to Meta, it is a step toward building personal superintelligence. The model is available on Meta AI and the Meta AI app. There is a private API preview available for selected users. Also Read: Meta layoffs continue in 2026, around 200 jobs cut in California

What is Muse Family and Muse Spark

To understand Muse Spark, we first need to understand what Muse Family is. Meta’s Muse Family is a new series of AI models developed by Meta Superintelligence Labs. This is a specialized division, which is led by Chief AI Officer Alexandr Wang. Muse Spark is the initial model to debut inside the Muse Family. The model is compact and fast and combines multiple abilities. Also Read: Meta’s new Ray-Ban AI Glasses now support prescription lenses and new features

The model supports visual chain-of-thought reasoning, multi agent orchestration, and tool use. The interesting thing about this model is that it can use several agents at the same time to work on tasks in parallel. Meta calls this mode as the ‘Contemplating Mode.’ Also Read: Meta’s next Ray-Ban smart glasses may be built for everyday users with even prescription

Availability

It will soon be rolled out to WhatsApp, Instagram, Facebook, Messenger, and AI Glasses in the coming weeks.

Today we’re introducing Muse Spark, our most powerful model yet, giving you a faster and smarter Meta AI.

Muse Spark currently powers the Meta AI app and website and will be rolling out to @whatsapp, @Instagram, @facebook, @messenger, and AI glasses in the coming weeks.…

— Meta Newsroom (@MetaNewsroom) April 8, 2026

As per Meta, this will improve performance on complex problems without slowing down responses. Besides this, Muse Spark can analyze objects and create interactive tutorials. It can also provide health-related guidance. The company worked with over 1000 physicians to improve the model’s health advice guidance.

Meta Spark vs Other AI Models

Meta performed evaluation of Muse Spark and compared it with other AI models and the results were mind blowing. As per Meta’s evaluation, Muse Spark outperformed other AI models like Gemini 3.1 Pro, GPT 54, and Grok 4.2 on many benchmarks.

Meta also compared the models on CharXiv reasoning, which is designed to test the ability of Multimodal Large Language Models. Muse Spark scored 86.4, which is higher than GPT 5.4 at 82.8. Gemini scored 80.2 and Grok received 60.9.

Besides this, Meta also conducted comparison on SimpleVQA that measures the ability for language models to answer short, fact-seeking questions. Under this measurement, Muse Spark scored 71.3, in ZeroBench, it scored 33.0, beating Gemini’s 29.0 but below GPT-5.4’s 41.0.

The company also carried out health tests too, wherein Muse Spark scored 42.8 on HealthBench Hard, placing it ahead of Gemini and Grok. This is also slightly above GPT 5.4.

Technology Used in Muse Spark

Meta’s new Muse Spark is built on a redesigned AI stack. This helps the model to improve its pretraining reinforcement learning and test-time reasoning.

Add Techlusive as a Preferred Source

Muse Spark Usage and Its Applications

Muse Spark can be used in practical applications, such as it can help you with education, health advice, and complex problems. It finally opens the door to more intelligent AI assistants. In future, the tech giant plans to develop the Muse Family by introducing more advanced models. Additionally, the company will also support to large-scale infrastructure like the Hyperion data center.