Liquid AI Releases LFM2.5 Models With Top Japanese Performance
The release gives Japanese-market developers a compact, open-weight model that claims leading benchmark scores for its size, potentially accelerating on-device AI agents in automotive, manufacturing, and smart devices.
Reporting from 1 sources: ASCII.jp.
Liquid AI has released the LFM2.5 family of on-device AI models, including a 1.2-billion-parameter Japanese language model and a 1.5-billion-parameter Japanese audio model. The company says the language model achieves top-tier performance among same-scale models on Japanese benchmarks, and the audio model handles speech recognition and dialogue without separate ASR or TTS components. The models are available as open weights on Hugging Face.
Liquid AI, a startup spun out of MIT, announced the LFM2.5 family on Thursday. The lineup includes the LFM2.5-1.2B-JP-202606, a 1.2-billion-parameter language model, and the LFM2.5-Audio-1.5B-JP, a 1.5-billion-parameter audio model. Both are designed for edge deployment. The company stated the language model outperforms comparable models across Japanese knowledge, instruction following, mathematics, coding, and tool use. The audio model is end-to-end, eliminating separate speech recognition and text-to-speech components. The models are released as open weights under the LFM Open License v1.0 and are available on Hugging Face. Liquid AI is also working with AMD, Qualcomm, and Nexa AI to deploy the models on NPUs.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.