LFM2.5‑VL-450M is Liquid AI's refreshed version of the first vision-language model, LFM2-VL-450M, built on an updated backbone LFM2.5-350M and tuned for stronger real-world performance. Find more about LFM2.5 family of models in our blog post.
Enhanced instruction following on vision and language tasks.
Improved multilingual vision understanding in Arabic, Chinese, French, German, Japanese, Korean, Portuguese and Spanish.
Bounding box prediction and object detection for grounded visual understanding.
Function calling support for text-only input.
🎥⚡️ You can try LFM2.5-VL-450M running locally in your browser with our real-time video stream captioning WebGPU demo 🎥⚡️
Alternatively, try the API model on the Playground.
We recommend using it for general vision-language workloads, captioning and object detection. It’s not well-suited for knowledge-intensive tasks or fine-grained OCR.
<|startoftext|><|im_start|>system
You are a helpful multimodal assistant by Liquid AI.<|im_end|>
<|im_start|>user
<image>Describe this image.<|im_end|>
<|im_start|>assistant
This image shows a Caenorhabditis elegans (C. elegans) nematode.<|im_end|>
This image captures a dramatic scene of an American flag prominently displayed on a flagpole against a clear, cloudless blue sky. The flag, with its iconic red and white stripes and blue field adorned with white stars, is fluttering in the wind, suggesting a strong breeze. The flagpole, which is black and topped with a spherical finial, supports the flag at an angle, with the flag fluttering to the left.
In the background, a rocket is seen launching into the sky, its nose pointed upwards and emitting a bright, white flame. The rocket's exhaust trails a vivid orange and yellow, creating a striking contrast against the clear blue sky. The image is taken from a low angle, emphasizing the flag's movement and the rocket's ascent, capturing a moment of national pride and technological achievement.
LFM2.5-VL-450M improves over LFM2-VL-450M across both vision and language benchmarks, while also adding two new capabilities: bounding box prediction on RefCOCO-M and function calling support measured by BFCLv4.
All vision benchmark scores are obtained using VLMEvalKit. Multilingual scores are based on the average of benchmarks translated by GPT-4.1-mini from English to Arabic, Chinese, French, German, Japanese, Korean, Portuguese, and Spanish.
# custom terms: read the license before you download$ fw get onnx-community/LFM2.5-VL-450M-ONNX
# manifest sha256 53eebd696f82… · 4.46 GB · 39 files$ fw verify ./LFM2.5-VL-450M-ONNX onnx-community/LFM2.5-VL-450M-ONNX
Terms that come with these files. LFM 1.0: commercial use only for organisations under US$10M annual revenue. Keep a copy of the license with the files.
Custom license (custom).Read the license text first. Open this panel to show the commands once you accept its terms.
fw.pyChecks the signature and every sha256, tries each source in order, resumes.