Liquid AI Models Now Run Much Faster

The company launched a feature called DSpark for its LFM2.5 model family. These updates act as a shortcut for the computer, letting it guess the next words before doing the heavy math. This makes the models feel much more responsive for users.

You can see these speed boosts on both powerful H100 servers and standard Apple MacBooks. The performance gain reaches over three times the speed on high-end hardware. Because the output stays the same, you get the same smart answers in a fraction of the time.

This release covers three different models including the 1.2B, 2.6B, and 8B versions. By adding a small amount of extra data, Liquid AI has managed to cut down wait times significantly. It is a big step forward for running AI directly on your own device.