Liquid AI reveals 8B-A1B MoE trained on 38T

In three linesLiquid AI unveils 8B-A1B, a Mixture of Experts model trained on 38 trillion tokens. The model combines 8 billion dense parameters with a modular expert architecture to optimize computational efficiency.

Summary generated by Claude — human-verified