LFM2.5 8B-A1B

Name: LFM2.5 8B-A1B
Author: Liquid AI

Liquid AI · LFM2.5 · released 2026-05-28 · data checked 2026-11-01

LFM2.5 8B-A1B is a 8.5B-parameter Mixture-of-Experts model (1B active per token) from Liquid AI (United States), released 2026-05-28 under the LFM Open License license. It doesn't have enough reported benchmarks for an index score yet. It needs about 5.9 GB of memory at 4-bit with an 8K-token context.

Small LM Index

—

Needs 2 core benchmarks (has 1)

Parameters

8.5B

1B active · MoE

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

128K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best small model on that benchmark. Sources: 6 reported by Liquid AI. How we label sources.

AIME Lab-reported Math · AIME 202542.5

Class best: Apriel 1.6 15B Thinker 88.0

τ²-Bench Lab-reported Agentic · Average of Telecom (88.1) and Retail (39.8)63.9

Class best: Qwen3.5 9B 79.1

IFBench Lab-reported Instruction following56.5

Class best: Granite 4.2 8B 79.3

IFEval Lab-reported Instruction following91.8

BFCL Lab-reported Agentic · BFCL v448.5

Class best: Hunyuan 7B Instruct 70.8

MATH-500 Lab-reported Math88.8

Class best: Falcon H1R 7B 97.4

Specifications

Lab: Liquid AI (US · United States)
Architecture: Mixture-of-Experts
Total parameters: 8.5B
Active parameters: 1B
Size class: Small
Context window: 128K tokens
Input modalities: text
Reasoning: Instruct (no thinking)
License: LFM Open License
Openness: Open weights
Memory to run: ~5.9 GB at 4-bit, 8K context (more sizes)
Hugging Face: 30.8K downloads (30 days) · 805 likes

Weights on Hugging Face ↗

Compare with

LFM2.5 8B-A1B vs DeepSeek-R1-0528-Qwen3-8B (8.2B, index 53.1)
LFM2.5 8B-A1B vs Qwen3 8B (8.2B, index 48.5)
LFM2.5 8B-A1B vs Nemotron Nano 9B v2 (8.9B, index 52.5)

Sources

https://huggingface.co/LiquidAI/LFM2.5-8B-A1B

Something wrong here? Report it.

Run it locally

Uses LiquidAI/LFM2.5-8B-A1B-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/LFM2.5-8B-A1B-GGUF:Q4_K_M

llama.cpp

llama-server -hf LiquidAI/LFM2.5-8B-A1B-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/LFM2.5-8B-A1B-MLX-4bit

vLLM (GPU server)

vllm serve LiquidAI/LFM2.5-8B-A1B

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	128K context
16-bit F16	18 GB	18 GB	19 GB
8-bit Q8_0	9.7 GB	10 GB	11 GB
4-bit Q4_K_M	5.9 GB	6.2 GB	7.4 GB

Check it against your GPU, Mac or phone →