LFM2.5-VL 3B

Name: LFM2.5-VL 3B
Author: Liquid AI

Liquid AI · LFM2.5 · released 2026-08-11 · data checked 2026-11-01

LFM2.5-VL 3B is a 3.1B-parameter dense model from Liquid AI (United States), released 2026-08-11 under the LFM Open License license. It doesn't have enough reported benchmarks for an index score yet. It needs about 2.6 GB of memory at 4-bit with an 8K-token context.

Small LM Index

—

Needs 2 core benchmarks (has 0)

Parameters

3.1B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

32K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 2 reported by Liquid AI. How we label sources.

MMMU-Pro Lab-reported Vision30.5

Class best: Qwen3.5 4B 65.4

BFCL Lab-reported Agentic · BFCL v432.5

Class best: Spark-X2.6 4B 71.8

Specifications

Lab: Liquid AI (US · United States)
Architecture: Dense transformer
Total parameters: 3.1B
Active parameters: All (dense)
Size class: Tiny
Context window: 32K tokens
Input modalities: text, image
Reasoning: Instruct (no thinking)
License: LFM Open License
Openness: Open weights
Memory to run: ~2.6 GB at 4-bit, 8K context (more sizes)
Hugging Face: 31.2K downloads (30 days) · 236 likes

Weights on Hugging Face ↗

Compare with

LFM2.5-VL 3B vs SmolLM3 3B (3.1B, index 30.4)
LFM2.5-VL 3B vs Granite 4.2 3B (3B, index 54.8)
LFM2.5-VL 3B vs Ministral 3 3B (3B, index 49.4)

Sources

https://huggingface.co/LiquidAI/LFM2.5-VL-3B

Something wrong here? Report it.

Run it locally

Uses LiquidAI/LFM2.5-VL-3B-GGUF (Q4_K_M). Swap the tag for another size: BF16, F16, Q4_0, Q4_K_M, Q5_K_M, Q6_K, Q8_0.

Ollama

ollama run hf.co/LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M

llama.cpp

llama-server -hf LiquidAI/LFM2.5-VL-3B-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model LiquidAI/LFM2.5-VL-3B-MLX-4bit

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve LiquidAI/LFM2.5-VL-3B

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context
16-bit F16	6.9 GB	7.3 GB
8-bit Q8_0	3.9 GB	4.4 GB
4-bit Q4_K_M	2.5 GB	3.0 GB

Check it against your GPU, Mac or phone →