LFM2.5 1.2B Instruct

Name: LFM2.5 1.2B Instruct
Author: Liquid AI

Liquid AI · LFM2.5 · released 2026-01-06 · data checked 2026-11-01

LFM2.5 1.2B Instruct is a 1.2B-parameter dense model from Liquid AI (United States), released 2026-01-06 under the LFM Open License license. It scores 20.6 on the Small LM Index, #17 of 21 tiny models. It needs about 1.3 GB of memory at 4-bit with an 8K-token context.

Hybrid convolution/attention architecture for on-device inference.

Small LM Index

20.6

#17 of 21 scored tiny models

Likely 17.7–23.9 · from 3 of 6 core benchmarks

Parameters

1.2B

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: 6 reported by Liquid AI. How we label sources.

MMLU-Pro Lab-reported Knowledge44.4

Class best: Qwen3.5 4B 79.1

GPQA Diamond Lab-reported Knowledge38.9

Class best: Nanbeige4.2 3B 87.4

AIME Lab-reported Math · AIME 202514.0

Class best: Spark-X2.6 4B 91.2

IFBench Lab-reported Instruction following47.3

Class best: Spark-X2.6 4B 75.8

IFEval Lab-reported Instruction following86.2

Class best: LFM3 1.7B 96.7

BFCL Lab-reported Agentic · BFCL v349.1

Class best: Spark-X2.6 4B 71.8

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 1

LiveCodeBench≈ 20

SWE-bench Verified≈ 6

Specifications

Lab: Liquid AI (US · United States)
Architecture: Dense transformer
Total parameters: 1.2B
Active parameters: All (dense)
Size class: Tiny
Context window: 32K tokens
Input modalities: text
Reasoning: Instruct (no thinking)
License: LFM Open License
Openness: Open weights
Memory to run: ~1.3 GB at 4-bit, 8K context (more sizes)
Hugging Face: 122.7K downloads (30 days) · 684 likes

Weights on Hugging Face ↗

Compare with

LFM2.5 1.2B Instruct vs EXAONE 4.0 1.2B (1.3B, index 36.1)
LFM2.5 1.2B Instruct vs LFM2.5 1.2B Thinking (1.2B, index 30.1)
LFM2.5 1.2B Instruct vs Gemma 3 1B (1B, index 6.1)
LFM2.5 1.2B Instruct vs Qwen3.5 0.8B (900M, index 10.7)
LFM2.5 1.2B Instruct vs LFM2.5 350M (350M, index 12.6)

Sources

https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct

Something wrong here? Report it.

Run it locally

Uses LiquidAI/LFM2.5-1.2B-Instruct-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/LFM2.5-1.2B-Instruct-GGUF:Q4_K_M

llama.cpp

llama-server -hf LiquidAI/LFM2.5-1.2B-Instruct-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/LFM2.5-1.2B-Instruct-MLX-4bit

vLLM (GPU server)

vllm serve LiquidAI/LFM2.5-1.2B-Instruct

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	2.9 GB	3.3 GB
8-bit Q8_0	1.8 GB	2.2 GB
4-bit Q4_K_M	1.3 GB	1.6 GB

Check it against your GPU, Mac or phone →