Meta bets big on ‘superintelligence personal’ – ditching llama, launching muse spark
Meta is throwing a considerable amount of its resources – and its reputation – behind a new AI initiative, Muse Spark, effectively abandoning its previous Llama project. The company’s top brass are betting the farm on this multimodal model to challenge the likes of GPT-5.4 and Gemini 3.1 Pro, a move that’s already drawing scrutiny.
A new contender in the generative ai arena
Initially, Meta’s attempts with generative AI – spearheaded by Llama – have been underwhelming, prompting a massive $4 billion investment in the Superintelligence Labs. Now, Muse Spark represents their first tangible product, a gamble that’s being framed internally as a direct competitor to OpenAI’s GPT and Google’s Gemini. Early benchmarks, according to Meta’s own data, show Muse Spark performing competitively in areas like multimodal perception, reasoning, health, and the development of AI agents – even surpassing Grok 4.2 and Opus 4.6.
However, the company isn’t painting a completely rosy picture. They acknowledge significant ‘performance deficiencies’ particularly when it comes to long-term agent behavior and code generation. This suggests a critical gap remains in the model's capabilities, a hurdle they’re actively working to overcome. Muse Spark’s architecture is built around a native multimodal reasoning engine, supporting tool usage, visual thought chains, and agent coordination – a concept Meta calls ‘superintelligence personal’.
The core functionality mirrors Gemini’s, accepting prompts through voice, text, and images, integrating seamlessly with Meta’s expanding suite of applications. Currently, access is limited to the Meta.ai web app and the main Meta app, but a wider rollout, including integration with WhatsApp, Instagram, Facebook, Messenger, and even Ray-Ban Meta smart glasses, is slated for the coming weeks. But there's more – Meta is developing a ‘Contemplating Mode,’ leveraging multiple AI agents working in parallel to achieve a level of reasoning previously only seen in models like Gemini Deep Think and GPT Pro. This feature, however, won’t be available until next month.
Adding another layer of complexity, Meta is pushing aggressively into the health sector – a notoriously sensitive area rife with privacy concerns. They’ve partnered with over 1,000 physicians to gather training data, aiming to deliver more objective and comprehensive responses. Muse Spark’s visualizations, dissecting health information from nutritional content to muscle activation during exercise, are intended to be a key differentiator. Yet, the potential for misuse – and the spread of unverified medical advice – remains a serious challenge.
Meanwhile, Anthropic’s impending release of Claude Mythos is generating alarm. Security researchers have identified thousands of critical vulnerabilities across operating systems and web browsers, highlighting the potential risks associated with this new AI. Meta’s reliance on multimodal capabilities, especially within its wearable Technology – spearheaded by the Ray-Ban Meta glasses – underscores the stakes. The ability to converse with AI via voice and interpret visual input is paramount to the success of this product, and Muse Spark is positioned to be the engine driving that functionality. Essentially, Meta is rebuilding its AI strategy from the ground up, shifting away from Llama and embracing Muse Spark as a modular, expanding competitor to GPT and Gemini.
The company’s commitment to this ambitious project is clear: they’re doubling down on a fundamentally different approach to artificial intelligence. And frankly, that's a bet nobody is going to take lightly.