Liquid AI Released LFM2.5-350M: A Compact 350M Parameter Model Trained on 28T Tokens with Scaled Reinforcement Learning

Liquid AI's LFM2.5-350M is a 350-million-parameter model.

It was pre-trained on 28 trillion tokens and trained using large-scale reinforcement learning.

The hybrid LIV backbone includes 10 Double-Gated LIV Convolution Blocks and 6 Grouped Query Attention blocks.

Supports a 32k (32,768-token) context window and is claimed to outperform models more than twice its size on several benchmarks.

Reported benchmark scores: IFEval (Instruction Following) 76.96; GPQA Diamond 30.64; MMLU-Pro 20.01.