LFM2.5-VL: Articles and Variants

September 25, 2026September 26, 2026

About This Model

LFM2.5-VL-3B is a small model that can also read images, from Liquid AI. The company’s LFM2.5 is a family of “hybrid" models designed to run fast on-device. LFM2.5-VL-3B pairs the LFM2.5-2.6B language model with an image encoder (SigLIP2 NaFlex, 400M), supports 16 languages including Japanese, and has a 32,768-token context window.

Compared with the previous generation (LFM2-VL-3B), it improves:

• Grounding objects from a natural-language query (for example, answering “where is the red car?" with coordinates)
• Full-page text recognition with layout information

What Makes It Stand Out

1. It is small and fast. In the publisher’s measurements it runs at 228 tokens/s on an Apple M5 Max and 116 tokens/s on an AMD Ryzen AI Max+ 395, in under 3.3GB of memory. You can use an image-reading model at practical speed without a big GPU machine.

2. A companion model, DSpark, makes it faster without changing the output. LFM2.5-VL-3B-DSpark, covered in an article on this site, is a small 279.5M-parameter draft model. It predicts several tokens ahead and the main model verifies them in one pass (speculative decoding), speeding up generation while producing the same output the main model would on its own (identical under greedy decoding; under matched sampling settings, the output distribution is preserved). The publisher’s measured speedups (decoding speed):

Tasks with predictable output, like captioning, speed up the most; less predictable ones, like conversation, gain less. End to end (including image processing), the speedups are smaller than these figures.

Its limits are also clearly stated. The publisher recommends single-turn, speed-sensitive tasks such as real-time object detection, batch OCR of scanned documents, and translating menus or road signs, and advises against long-context or reasoning-heavy tasks, such as highly technical questions about blueprints.

Running It Locally

• The article on this page and the “Our Coverage and Data" section below cover DSpark, the draft model, not the main model. DSpark does not run on its own; it is used together with LFM2.5-VL-3B.
• The publisher distributes GGUF (for llama.cpp), ONNX and MLX (for Mac) builds.
• To use DSpark you need SGLang v0.5.19 or later, MLX-VLM v0.7.2 or later (currently --temperature 0 only), or llama.cpp with the DSpark GGUF build.
• The license is Liquid AI’s own LFM Open License v1.0. Commercial use by businesses above its US$10 million annual-revenue threshold is not licensed under it (commercial use below the threshold is allowed). Check the original terms before using it for work.

Sources: model cards and license for LiquidAI/LFM2.5-VL-3B and LiquidAI/LFM2.5-VL-3B-DSpark, as of 2026-09-25. Speed figures are the publisher’s measurements.

Our Coverage and Data

Everything Local Model Watch has published about the LFM2.5-VL family: 1 article(s) covering the base model and its fine-tunes, plus converted builds we tracked after publication. Memory requirements below are computed by this site from file sizes, not quoted from model cards. Part of our model family index.

At a Glance

Quantized and Converted Variants

File sizes of each build:

• Available builds in LiquidAI/LFM2.5-VL-3B-DSpark-GGUF: F16 0.5GB

This section is appended automatically by Local Model Watch when a converted build of this model appears after publication. Memory figures are estimated from the size of the distributed files.

Articles (the family’s own models first, then newest)

Repositories

• LiquidAI/LFM2.5-VL-3B
• LiquidAI/LFM2.5-VL-3B-DSpark

Last updated 2026-09-26 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log.