Liquid AI has released LFM2.5-VL-3B, a 3.1‑billion‑parameter on-device vision-language model that reads screens, grounds objects, parses documents and charts, supports multi-image input, and performs tool calling from text or images. Built on an LFM2.5 language backbone with a SigLIP2 NaFlex vision encoder, the model supports 16 languages, a 32,768-token context, and scores 69.4 across 28 benchmarks, matching larger 4.7B-parameter competitors. Fitting in roughly 3 GB and decoding quickly on consumer hardware, the model targets applications from GUI agents to automotive perception. A revenue-tiered open license lets smaller companies deploy commercially while larger enterprises negotiate paid terms.
This update represents a notable development in the Ai sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.