← All models

LFM2-VL-450M

Liquid AI · 2025-08-12 · 451M parameters

LFM2-VL-450M is the smaller of Liquid AI's first two vision-language models, released August 12, 2025 for highly constrained devices: the LFM2-350M language model with an 86M-parameter SigLIP2 NaFlex base vision encoder that reads images at native resolution. The card gives a 32,768-token text context and English support. The BF16 checkpoint contains 450,822,656 parameters, and the tokenizer assigns 64,400 distinct IDs within 65,536 embedding rows.

Benchmark scores