Meta Debuts Muse Spark as First Model from New Superintelligence Labs

Meta Platforms unveiled Muse Spark on Wednesday, its first artificial intelligence model produced by the newly formed Meta Superintelligence Labs, marking a significant departure from the company's previous open-source Llama strategy. The model is natively multimodal, meaning it processes text, images, and video simultaneously without relying on separate modules, and is part of an internal family of models codenamed Avocado. Muse Spark is currently available on the Meta AI app and the meta.ai website, and the company said it will replace existing Llama-powered chatbots on WhatsApp, Instagram, Facebook, Messenger, and Meta's smart glasses in the coming weeks. The launch comes after Mark Zuckerberg spent billions overhauling Meta's AI program following the delayed and disappointing release of Llama 4 in 2025, which was widely viewed in the tech industry as a middling performer.

Meta's previous flagship AI series, Llama, was released as open-source software, making it available for researchers, startups, and developers to download and customize. Llama 4 launched in April 2025 but failed to meet internal expectations, particularly on reasoning and mathematics, according to reporting by Wired and Il Sole 24 ORE. The underwhelming reception prompted Zuckerberg to reorganize Meta's AI program and establish Meta Superintelligence Labs as a dedicated unit pursuing what the company calls "personal superintelligence." Meta recruited Alexandr Wang, co-founder and former chief executive of data labeling company Scale AI, to lead the new division as Chief AI Officer, alongside engineers poached from OpenAI, Anthropic, and Google.

The release represents a strategic pivot for Meta, which had long been regarded as a leader in open-source AI but is now keeping Muse Spark closed-source, at least initially. The company said it hopes to open-source future versions of the model. Muse Spark introduces a feature called Contemplating mode, which runs multiple AI agents in parallel to tackle complex reasoning problems, positioning it as a direct competitor to Google's Gemini Deep Think and OpenAI's GPT Pro extended thinking modes. Meta stated that Muse Spark can match the performance of Llama 4 Maverick while using ten times less computing power, a claim that addresses the enormous energy and infrastructure costs associated with generative AI. The company is betting that integrating the model across its platforms will give it a distribution advantage over rivals with smaller user bases. Meta also said it collaborated with over 1,000 physicians to curate training data aimed at improving the model's health reasoning capabilities, a feature the company is emphasizing alongside a new shopping mode that draws on creator content from Instagram and Facebook.

Benchmark scores put Muse Spark in top five globally

Artificial Analysis, an AI benchmarking firm that received early access to the model, said Muse Spark scored 52 on its Intelligence Index, placing it within the top five models it has tested. „Muse Spark scores 52 on the Artificial Analysis Intelligence Index, placing it within the top 5 models we have benchmarked” — Artificial Analysis via Wired Meta's own benchmark data showed the model performing significantly better than previous Meta models on writing and reasoning tasks, and nearly on par with top offerings from Google, OpenAI, and Anthropic, though the company acknowledged it still has gaps in coding ability and long-horizon agentic workflows. Gizmodo noted that Meta has previously been accused of presenting favorable benchmark figures without releasing an accompanying technical paper, and advised readers to treat the numbers with caution. The New York Times reported that Muse Spark's coding performance lags rivals, particularly Anthropic, which has become the dominant player in AI-assisted coding. Meta's announcement came one day after Anthropic said its latest model, Mythos, was too powerful to safely release due to cybersecurity concerns.

Wang's $14.3 billion bet faces its first public test

The launch is the first tangible output of Meta's Scale AI investment, through which Meta paid $14.3 billion for a 49% stake in the data labeling firm and recruited Wang to lead its superintelligence effort. Alexandr Wang, 29, now serves as Meta's Chief AI Officer and oversees the Superintelligence Labs responsible for Muse Spark. Zuckerberg wrote on Threads that Meta's goal is to build AI products that act as agents performing tasks on behalf of users, not merely answering questions. „Looking ahead, we plan to release increasingly advanced models that push the frontier of intelligence and capabilities, including new open source models” — Mark Zuckerberg via Wired Meta said Muse Spark is described as only the first step on its "scaling ladder," with the next generation of models already in development. The company did not disclose the model's parameter size, a standard measure used to compare AI systems' computing requirements, making independent verification of its efficiency claims difficult. Privacy concerns have also been raised, as users must log in with an existing Facebook or Instagram account to access Muse Spark, and Meta has not explicitly stated whether personal account data will be used to train or personalize the model.

Muse Spark: 52

49% (stake) — Meta's ownership share in Scale AI