Meta Launches Muse Spark AI After Llama 4 Stumble Following the chilly reception to its Llama 4 models, Meta is pivoting with a new AI family from its Superintelligence team. The first release is Muse Spark, a lightweight model designed for everyday users, marking the start of what Meta calls its Muse era. The company aims to master fundamental capabilities before moving to more advanced versions. Spark enters a competitive field with features now considered standard for 2026. It offers two response modes: Instant for quick replies and Thinking, which takes extra time to reason through complex prompts. This hybrid approach mirrors systems like Anthropic’s Claude Sonnet 3.7 from last year. Meta plans to add an even more powerful Contemplating mode in the future. A key feature is Spark’s ability to coordinate multiple AI subagents for a single task. For example, when planning a family trip, one agent could build an itinerary while another finds suitable activities for children. The model is also natively multimodal, processing images, video, and audio. Users can take a photo with their phone and ask Meta AI questions about it, similar to functionality offered by Google Lens. True to current trends, Spark includes a built-in shopping assistant. Like ChatGPT’s offering, it can compare products, list pros and cons, and provide links for purchase. Muse Spark is available now in the Meta AI app and on the meta.ai website. The new features are rolling out first in the US, with expansion to more countries and integration into Facebook, Instagram, and WhatsApp planned in the coming weeks. In a note on openness, Meta stated it hopes to open source future versions of the model. This follows apparent shifts in the company’s stance last year, when CEO Mark Zuckerberg indicated a more rigorous approach to such decisions.
