Meta Platforms has debuted its first major artificial intelligence model after the recent hiring of Alexandr Wang nine months ago. Originally codenamed Avocado, the new development has now been dubbed Muse Spark. The new AI model, announced on Wednesday, will be the first from the company’s new Muse series.
Meta is pursuing a rebound in the competitive AI market following the disappointing debut of its latest open-source models last April. The release failed to captivate developers, leading CEO Mark Zuckerberg to pivot the company’s strategy. Meta has been highlighting its efficiency and competitive performance on various tasks.
Meta highlights that Muse Spark delivers competitive performance across multimodal perception, reasoning, health, and agentic tasks. The company has pointed out that it continues to invest in areas with performance gaps, such as long-horizon agentic systems and coding workflows. With the launch, Meta has also introduced a “contemplation mode” that enables multiple agents to reason in parallel. This mode, as per Meta, competes closely with the extreme reasoning modes of frontier models such as Gemini Deep Think and GPT Pro.
Newer applications in different fields
Muse Spark is a superintelligence that is aware of its surroundings and understands the user’s world. It can analyze environments and support wellness. Meta details how Muse Spark’s advanced reasoning capabilities enable powerful, highly personal use cases.
Multimodal: As per Meta, Muse Spark has been built from the ground up to integrate visual information across domains. It achieves strong performance on visual STEM questions, entity recognition, and localization. These capabilities help in creating interactive applications such as minigames or troubleshooting home appliances.
To further improve Muse Spark’s health reasoning capabilities, the model has been trained on data curated by 1,000 physicians. Muse Spark can unpack and explain health information, such as the nutritional content of various foods. The contemplation mode provides significant capability improvements on challenging tasks, achieving 58% on Humanity’s Last Exam and 38% on Frontier Science Research. better than Gemini 3.1 deep think and GPT 5.4 pro.
Safety checks
Muse Spark has been extensively evaluated for safety across dual-use scientific domains. The process followed the updated Advanced AI Scaling Framework, which defines threat models, evaluation protocols, and deployment thresholds for advanced models. The tests have revealed that Muse Spark exhibits strong refusal behavior across high-risk domains, including biological and chemical weapons. This is enabled by pretraining data filtering, safety-focused post-training, and system-level guardrails.
Muse Spark does not exhibit the autonomous capability or hazardous tendencies needed to realize threat scenarios.
Meta regards Muse Spark as a first step on its scaling ladder and the first product of a ground-up overhaul of its AI efforts.
