What is Muse Spark? Meta’s newly launched AI model and how it compares to ChatGPT and Claude

AI companies are launching new models at a rapid pace. In the latest development, Meta has introduced a new AI model called Muse Spark, which the company says is its most advanced system so far. The model was developed by Meta Superintelligence Labs (MSL), a new research division set up by Meta in 2025. The team is led by Alexandr Wang, who joined the company after Meta acquired his startup Scale AI in a multi-billion-dollar deal.

According to company announcements and developer discussions, Muse Spark is designed to power Meta’s AI assistant across platforms such as Facebook, Instagram, and WhatsApp.

What is Meta Muse Spark?

Muse Spark is the first model from Meta’s new Muse family of AI systems, which is expected to replace the company’s earlier Llama models over time. Unlike some large flagship models, Muse Spark is described as a small but fast AI model. Meta says the model was built from scratch in about nine months.

Despite its smaller size, the company claims the model can still handle complex reasoning tasks. In its technical description, Meta said the system can process questions related to science, mathematics and health.

According to Alexandr Wang, the model also includes a “contemplating mode”, where multiple AI agents can work in parallel to solve complex problems.

How Muse Spark compares with ChatGPT and Claude

Muse Spark enters a highly competitive market where several companies are developing advanced AI assistants.

Major competitors include:

ChatGPT developed by OpenAI

Claude developed by Anthropic

Meta claims that Muse Spark performs competitively in certain benchmarks, including:

Multimodal reasoning

Scientific problem solving

Health-related analysis

Agent-based workflows

The company also said early testing showed the model could compete with other advanced reasoning systems such as Google’s Deep Think models and OpenAI’s professional reasoning models.

However, these comparisons are based on internal tests and have not yet been widely verified by independent researchers.

What can Muse Spark do?

Meta says the new AI system is designed to understand real-world context and assist users across its ecosystem of apps.

Some of its capabilities include:

Solving science and mathematics problems

Answering complex questions

Managing multiple AI agents for tasks

Providing recommendations for shopping and creators

Understanding conversations across Meta apps

The model will power an updated version of Meta AI, which the company plans to integrate across Facebook, Instagram, WhatsApp, Messenger and Meta’s smart glasses.

Meta says the system is being built for the three billion people who use its platforms globally.

Muse Spark is not open source. One major difference from Meta’s earlier AI models is that Muse Spark is currently closed source.

Meta’s previous Llama models were released publicly, allowing developers to access the code and build applications. Muse Spark, however, is currently available only through a private API preview for selected partners. The company said future versions of the Muse models may be open sourced, but no timeline has been confirmed.

Safety testing and future development

Meta said the new AI model has undergone safety evaluations, particularly in areas such as biological or chemical misuse.

According to Alexandr Wang, Muse Spark showed strong refusal behaviour when asked about high-risk topics during internal testing.

The company added that larger AI models in the Muse family are already under development, with new infrastructure being built to support them.

This suggests Muse Spark may be only the first step in Meta’s long-term AI roadmap.

Why Muse Spark matters in the AI race

The launch of Muse Spark reflects the growing competition among technology companies to build powerful AI assistants. Companies including Meta, OpenAI, Anthropic and Google are investing billions of dollars into AI research and infrastructure. Meta’s strategy is different in one key way: the company plans to integrate AI deeply across its social platforms, where billions of users interact every day.

If Muse Spark performs as expected, it could play a major role in shaping how people use AI inside messaging apps, social media and digital services.