The new model has been developed by Meta Superintelligence Labs (MSL), marking a significant step in the company's long-term AI ambitions.

The launch also coincides with a refreshed identity for Meta AI, as the company looks to integrate more advanced capabilities across its ecosystem of apps and services.

The Muse Spark model is the first Large Language Model (LLM) built entirely by Meta Superintelligence Labs, an initiative spearheaded by Mark Zuckerberg. The lab is led by Alexandr Wang, who joined the company following Meta's reported investment of over ₹1.15 lakh crore (approximately $14 billion) in Scale AI.

According to Wang, Muse Spark was developed from scratch within nine months, underscoring the accelerated pace at which Meta is advancing its AI infrastructure. The model will now power Meta AI across platforms such as Facebook, Instagram and WhatsApp.

Muse Spark is believed to have been internally tested under the codename 'Avocado' before its official release.

Muse Spark is part of Meta's new 'Muse' family of AI models, which are expected to eventually replace the company's earlier Llama series. Unlike flagship models, Spark is designed to be smaller and faster while still delivering advanced reasoning capabilities.

Meta claims that despite its compact design, the model performs competitively with top-tier systems like Anthropic's Claude Opus 4.6 and OpenAI's GPT-5.4 in benchmarks involving reasoning, multimodal understanding, and health-related applications.

A notable feature is its 'contemplating mode', which allows the system to coordinate multiple AI agents simultaneously to solve complex problems. This capability places it in direct competition with advanced reasoning systems such as Gemini.

According to Meta, Muse Spark is designed to 'understand the world around you', enabling it to process real-world context alongside user inputs. The model can assist with scientific queries, mathematical problem-solving, and contextual recommendations based on user activity across Meta platforms.

The updated Meta AI powered by Muse Spark is currently available via a dedicated app and web interface, with plans to expand to Messenger and Meta's smart glasses in the near future.

Additionally, the company has introduced a shopping feature within Meta AI, allowing users to receive recommendations based on creators, brands, and their interactions across apps.

One of the most notable shifts in strategy is Meta's move away from open-source AI. While its earlier Llama models were publicly accessible, Muse Spark is a closed-source system. This means its underlying code is not available to developers or researchers at present.

However, Meta has indicated that future versions of the Muse models could potentially be open-sourced, depending on development and safety considerations.

Currently, Muse Spark is being rolled out as a private API preview to select partners. The company has confirmed that more advanced and larger models are already under development, supported by expanding infrastructure.

Meta has emphasised safety as a key focus area in the development of Muse Spark. According to Alexandr Wang, the model demonstrated strong refusal behaviour in high-risk scenarios, including queries related to biological and chemical threats, during internal testing.

This positions Muse Spark as not just a high-performance model but also one aligned with emerging AI safety standards.

With the launch of Muse Spark, Meta is making a decisive push into the next phase of artificial intelligence, directly challenging established players in the space. While still in its early stages, the model reflects Meta's ambition to integrate advanced AI deeply into its platforms used by billions worldwide. The coming months will determine whether Muse Spark can truly match or surpass its rivals in real-world performance and adoption.