Liquid AI Releases High-Performance Small Language and Audio Models for Japanese

Liquid AI, Inc., an MIT-originated startup (Co-founder & CEO: Ramin Hasani, HQ: Cambridge, MA), and its Japanese subsidiary, Liquid AI, have released new models 'LFM2.5-1.2B-JP-202606' and 'LFM2.5-Audio-1.5B-JP' for deployment on edge devices. This release supports the practical application of on-device AI agents across broad Japanese market sectors including automotive, manufacturing, robotics, finance, public services, and smart devices.

Japanese Language Model LFM2.5-1.2B-JP-202606 LFM2.5-1.2B-JP-202606 achieves peak performance for its size across a wide range of capabilities, including Japanese language knowledge, instruction following, mathematics, coding, and tool use.

Japanese Audio Model LFM2.5-Audio-1.5B-JP LFM2.5-Audio-1.5B-JP is an end-to-end multimodal speech-to-text language model that eliminates the need for independent ASR or TTS components. With only 1.5 billion parameters, it enables seamless Japanese conversational interactions.

The company has partnered with AMD, Qualcomm, and Nexa AI to deploy the LFM2.5 family for NPU environments. The models are available as open weights on Hugging Face under the LFM Open License v1.0.