Meta Platforms Inc. has released eight new AI research artifacts for non-commercial use, including Meta Spirit LM, its first open-source multimodal language model capable of freely mixing text and speech inputs and outputs. Unveiled at META FAIR, Meta's event focused on advancing machine intelligence (AMI), Spirit LM can detect and reflect emotional cues such as anger, surprise, and joy through its Expressive variant. Unlike existing AI voice systems that rely on separate speech recognition and text synthesis components, Meta Spirit LM integrates text and speech processing within a single model, enabling more natural and expressive interactions. Additionally, Meta released an updated Meta Segment Anything Model 2.1, offering improved results on visually similar objects, small objects, and occlusion handling. These releases underscore Meta's commitment to open science and accelerating innovation in artificial intelligence.