**Meta Introduces Muse Spark: A New Chapter in AI Models**
Meta has officially introduced Muse Spark, a new AI model that supersedes its earlier Llama models. This advancement signifies a notable transformation in Meta’s AI approach, which has been redefined under the guidance of Alexandr Wang and the newly established Meta Superintelligence Labs. Muse Spark is presented as the initial phase in a succession of models aimed at improving AI functionalities across various fields.
**Competitive Performance and Attributes**
Muse Spark is crafted to provide competitive performance in several critical domains, including multimodal perception, reasoning, health, and agentic tasks. Meta underscores its dedication to closing existing performance gaps, particularly in long-horizon agentic systems and coding workflows. The model has been evaluated against top competitors like Anthropic’s Claude Opus 4.6 Max, Google’s Gemini 3.1 Pro High, OpenAI’s GPT-5.4 Xhigh, and xAI’s Grok 4.2 Reasoning models, showing favorable outcomes.
Starting today, Muse Spark is accessible in Meta AI applications, enabling users to experience its features directly.
**Launching Contemplating Mode**
Alongside Muse Spark Thinking, Meta is poised to unveil a new function named Contemplating mode. This mode aims to rival the advanced reasoning abilities present in models like Gemini 3.1 Deep Think and GPT-5.4 Pro. Contemplating mode coordinates multiple agents that can reason simultaneously, greatly boosting the model’s effectiveness in intricate tasks. Meta reports that this mode has registered commendable scores in demanding evaluations, including 58% in Humanity’s Last Exam and 38% in FrontierScience Research.
A private API preview for selected partners will also be launched, broadening the application and reach of Muse Spark.
**Emphasis on Health and Multimodal Proficiency**
Muse Spark stands out in multimodal proficiency, especially in areas like visual STEM questions, entity recognition, and localization. A notable characteristic is its use in health-related assignments. Meta has partnered with over 1,000 physicians to assemble training data that strengthens Muse Spark’s health reasoning abilities. This collaboration allows the model to provide accurate and detailed responses, including generating interactive visuals that clarify health information, such as nutritional details and muscle activation during physical activity.
For those keen on delving deeper into the new Muse Spark AI models, further information is available on the Meta Superintelligence Labs website.
**Looking Forward**
As Meta continues to push boundaries in the AI realm, excitement grows for forthcoming advancements, including the outcomes of Apple’s partnership with Google, anticipated to be revealed at WWDC 2026. This event will highlight iOS 27, featuring Gemini-driven Siri and new Apple Intelligence functionalities, set to be revealed on June 8.