450M-class VLM (LFM2.5-350M text backbone + SigLIP2 NaFlex ~86M vision encoder)
128K tokens
LFM Open License v1.0
Open-weight refreshed 450M vision-language release built on the LFM2.5-350M backbone and positioned for stronger real-world edge multimodal performance than LFM2-VL-450M. Official model-card claims highlight improved instruction following on vision-language tasks, improved multilingual vision understanding across nine languages, and new/expanded grounded capabilities including bounding-box prediction and object detection, plus function-calling support for text-only prompts. Release documentation emphasizes practical on-device deployment and flexibility: native-resolution handling up to 512x512 without forced aspect-ratio distortion, tiling with thumbnail context for larger images, and tunable image-token budgets at inference for speed/quality trade-offs. Published benchmark tables report broad gains over LFM2-VL-450M across MMStar, RealWorldQA, MMBench, MM-IFEval, OCRBench, CountBench, RefCOCO-M, and language-side IFEval/Multi-IF/BFCLv4, with release availability across Transformers, GGUF/llama.cpp, ONNX, and MLX variants.