Meta Drops Muse Spark: A New Multimodal AI Bet on Personal Superintelligence

Meta has unveiled Muse Spark, the first model in its new Muse family developed by Meta Superintelligence Labs, marking a significant step in the company’s push toward advanced artificial intelligence systems.

The model, launched on Meta’s AI platform and app, is described as a natively multimodal reasoning system capable of handling text, images, and tool-based tasks while supporting multi-agent orchestration.

Muse Spark represents the first outcome of a broader overhaul of Meta’s AI infrastructure, alongside major investments in research, training systems, and computing capacity, including the Hyperion data centre.

"Muse Spark offers competitive performance in multimodal perception, reasoning, health, and agentic tasks. We continue to invest in areas with current performance gaps, such as long-horizon agentic systems and coding workflows," Meta said.

A key feature introduced alongside the model is “Contemplating mode,” which enables multiple AI agents to reason in parallel. This approach is designed to rival advanced reasoning systems from competing models and has shown improved results on complex benchmarks, including Humanity’s Last Exam and FrontierScience Research.

"We’re also releasing contemplating mode, which orchestrates multiple agents that reason in parallel, designed to handle complex scientific & reasoning queries. In our testing, we found it competitive with other extreme reasoning models such as Gemini Deep Think & GPT Pro," Alexandr Wang, Meta Chief AI Officer, said.

Meta emphasised Muse Spark’s potential for highly personalised applications. The model can interpret visual inputs to assist with tasks such as troubleshooting devices or generating interactive experiences.

In health-related use cases, it has been trained with curated data developed in collaboration with more than 1,000 physicians, enabling it to provide detailed explanations of nutrition, exercise, and wellness.

The company also highlighted advances in efficiency, claiming its updated training approach achieves similar performance levels with significantly reduced computational resources compared to earlier models.

Following internal and third-party safety evaluations, Meta stated that Muse Spark meets its deployment standards, demonstrating strong safeguards in high-risk domains.

"In third-party evaluations on a near-launch checkpoint, Apollo Research found that Muse Spark demonstrated the highest rate of evaluation awareness of models they have observed," Meta added.

A broader safety report is expected to be released soon.