Liquid AI released 4-bit quantized LFM2.5 checkpoints, trained with quantization-aware distillation, that keep 97% of full-precision quality with no runtime changes needed.
Continue to AI University →