Liquid AI Unveils LFM2.5-VL-3B: A Compact Vision-Language Model for On-Device Screen Reading, Object Grounding, and Tool Calling

Liquid AI announced the release of LFM2.5-VL-3B, a 3.1B-parameter vision-language model designed for on-device deployment. The model reads digital screens across mobile, web, and desktop, grounds objects to coordinates, parses documents and charts, and calls tools from text or image input. It averages 69.4 across 28 vision benchmarks, matching InternVL-3.5-4B and sitting 0.7 points behind Qwen3.5-4B, both of which are 4.7B models. The model is non-reasoning to keep latency low, fits in roughly 3 GB of memory, and decodes 228 tokens/s on an Apple M5 Max. The checkpoint is available in native, GGUF, ONNX, and MLX formats, with day-one support for llama.cpp, MLX, vLLM, SGLang, and ONNX runtimes. The LFM Open License v1.0 is Apache-2.0-based but free commercial use ends when a company's annual revenue reaches $10M. Key improvements over the previous version include a ScreenSpot-v2 average of 80.7, a RefCOCO-avg precision@1 increase from 57.1 to 87.9, and new function calling capabilities with ToolSandbox scores improving from 26.4 to 59.5.