LFM2-VL-3B - Liquid AI Docs

This model is deprecated. Use LFM2.5-VL-3B for improved performance.

LFM2-VL-3B was the largest model in the LFM2-VL series, built on the LFM2-2.6B language backbone with a SigLIP2 NaFlex vision encoder. It has been superseded by LFM2.5-VL-3B, which improves grounding, screen understanding, multi-image input, and tool use, and supports more languages.

HF GGUF MLX ONNX

Specifications

Property	Value
Parameters	3B
Context Length	32K tokens
Architecture	LFM2-VL (Dense)
Languages	English
License	LFM Open License v1.0

Visual Reasoning

Fine-grained perception and visual Q&A

OCR

Text recognition in images

Native Resolution

Up to 512×512 without upscaling, tiling for larger images

Quick Start

Transformers
vLLM
SGLang
llama.cpp