Liquid AI ships quantized LFM2.5 checkpoints

Liquid AI published Q4_0-quantized checkpoints of its LFM2.5 family, produced through quantization-aware distillation rather than post-training quantization, aimed at shrinking the on-device model footprint while limiting the usual accuracy loss from 4-bit quantization. It’s the latest in a steady cadence of LFM2.5 variants Liquid AI has shipped this summer, including an edge-focused vision model in August.