LFM2.5-VL-3B
Liquid AI · Vision-language · 3.1B · Released 12 August 2026
A 3.1-billion-parameter open-weight vision-language model from Liquid AI, built to run on-device: phones, laptops, and single consumer GPUs rather than in a data centre. It adds screen understanding, object grounding, multi-image reasoning, and function calling, and ships with day-one GGUF, MLX, and ONNX exports.
Strengths
- Small enough to run on a phone or any recent laptop
- Broad day-one runtime support across llama.cpp, MLX, vLLM, SGLang, and ONNX
- Screen understanding, object grounding, and function calling suit on-device agents
Weaknesses
- The licence restricts free commercial use by organisation revenue (see below)
- A 3B model trails larger VLMs on hard visual reasoning
- Vision support in local runtimes is less mature than for text-only models
Hardware requirements
| Quantisation | Approx. VRAM | Notes |
|---|---|---|
| Q4_K_M | ~3GB | Approximate, including the vision encoder; runs on a phone or laptop |
| Q8_0 | ~4GB | Approximate; comfortable on any recent machine |
| BF16 | ~7GB | Approximate; full precision |
Also runs on CPU (slower). Optimised builds available for Apple Silicon.
What you'd need to run this
Roughly what a machine to run this would need, at up to three levels of quality. Memory is the deciding factor.
Minimum to run it
Q4_K_M · ~3GB needed
One 12GB GPU
NVIDIA GeForce RTX 3060 12GBor a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
At least 32GB of system RAM alongside the card.
around £700–£1,100
What else 12GB runs →For good quality
Q8_0 · ~4GB needed
One 12GB GPU
NVIDIA GeForce RTX 3060 12GBor a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
At least 32GB of system RAM alongside the card.
around £700–£1,100
What else 12GB runs →Best quality
BF16 · ~7GB needed
One 12GB GPU
NVIDIA GeForce RTX 3060 12GBor a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
At least 32GB of system RAM alongside the card.
around £700–£1,100
What else 12GB runs →Licence
LFM Open License v1.0
LFM2.5-VL-3B: common questions
- What hardware do I need to run LFM2.5-VL-3B?
- At its most compressed (Q4_K_M) it needs roughly 3GB of VRAM, and about 4GB for good quality. VRAM figures are approximate and depend on context length and settings.
- Is LFM2.5-VL-3B free for commercial use?
- Commercial use is permitted, but with conditions. Liquid AI describes the licence as permitting commercial use free of charge only for organisations under 10 million US dollars in annual revenue; above that threshold a separate commercial arrangement with Liquid AI is required. Read the licence before relying on it at scale.
- Can I run LFM2.5-VL-3B on Apple Silicon?
- Yes. LFM2.5-VL-3B has builds optimised for Apple Silicon, through MLX or GGUF on a Mac.
- Does LFM2.5-VL-3B run on CPU?
- Yes, LFM2.5-VL-3B can run on the CPU, though generation is slower than on a GPU.
Availability
Recommended for
- On-device image, document, and screen understanding
- Privacy-sensitive or offline visual tasks
- On-device agents that need to see an interface and call tools
Related models
Glossary
Catalogue entry last verified 27 August 2026. Specifications change; verify anything you are about to spend money on.