LFM2.5 24B-A2B

Name: LFM2.5 24B-A2B
Price: 0.213 USD
Author: Liquid AI

Liquid AI · LFM2.5 · released 2026-10-13 · data checked 2026-11-01

LFM2.5 24B-A2B is a 24B-parameter Mixture-of-Experts model (2B active per token) from Liquid AI (United States), released 2026-10-13 under the LFM Open License license. It scores 61.4 on the Small LM Index, #19 of 34 medium models. It needs about 15.7 GB of memory at 4-bit with an 8K-token context and costs $0.21 per million tokens through the cheapest API provider.

Small LM Index

61.4

#19 of 34 scored medium models

Likely 57.3–65.2 · from 3 of 6 core benchmarks

Parameters

24B

2B active · MoE

Price per 1M tokens

$0.21

$0.050 in / $0.70 out · blended 3:1

Context window

128K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 8 reported by Liquid AI. How we label sources.

GPQA Diamond Lab-reported Knowledge74.4

Class best: Nanbeige4.3 33B 89.5

Humanity's Last Exam Lab-reported Reasoning7.4

Class best: Qwen3.8 27B 30.8

SWE-bench Verified Lab-reported Coding · OpenCode harness51.0

Class best: Qwen3.6 27B 77.2

IFBench Lab-reported Instruction following74.7

Class best: Sage-A1.5 81.6

IFEval Lab-reported Instruction following96.8

BFCL Lab-reported Agentic · BFCL v465.0

Class best: Ternary Bonsai 2 27B 74.9

MATH-500 Lab-reported Math96.0

Class best: Ternary Bonsai 2 27B 98.8

SciCode Lab-reported Coding32.7

Class best: Sage-A1.5 44.3

Not reported: estimated

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

MMLU-Pro≈ 79

AIME≈ 85

LiveCodeBench≈ 71

Specifications

Lab: Liquid AI (US · United States)
Architecture: Mixture-of-Experts
Total parameters: 24B
Active parameters: 2B
Size class: Medium
Context window: 128K tokens
Input modalities: text
Reasoning: Instruct (no thinking)
License: LFM Open License
Openness: Open weights
Memory to run: ~15.7 GB at 4-bit, 8K context (more sizes)
Hugging Face: 32.8K downloads (30 days) · 596 likes
API providers: 1 on OpenRouter
API price: $0.050 in / $0.70 out per 1M (OpenRouter, 2026-11-03)

Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

LFM2.5 24B-A2B vs Gemma 4 26B-A4B (25.2B, index 67.5)
LFM2.5 24B-A2B vs gpt-oss-20b (20.9B, index 64.0)
LFM2.5 24B-A2B vs Magistral Small 1.2 (24B, index 55.9)
LFM2.5 24B-A2B vs Mistral Small 3.2 (24B, index 38.6)
LFM2.5 24B-A2B vs EuroLLM 22B Instruct (22.6B, index 19.2)
LFM2.5 24B-A2B vs Tri-21B Think (20.7B, index 43.6)

Sources

https://huggingface.co/LiquidAI/LFM2.5-24B-A2B

Something wrong here? Report it.

Run it locally

Uses LiquidAI/LFM2.5-24B-A2B-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/LFM2.5-24B-A2B-GGUF:Q4_K_M

llama.cpp

llama-server -hf LiquidAI/LFM2.5-24B-A2B-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/LFM2.5-24B-A2B-MLX-4bit

vLLM (GPU server)

vllm serve LiquidAI/LFM2.5-24B-A2B

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	49 GB	50 GB	52 GB
8-bit Q8_0	27 GB	27 GB	29 GB
4-bit Q4_K_M	16 GB	16 GB	18 GB

Check it against your GPU, Mac or phone →