LFM2.5 2.6B

Name: LFM2.5 2.6B
Price: 0.387 USD
Author: Liquid AI

Liquid AI · LFM2.5 · released 2026-07-28 · data checked 2026-11-01

LFM2.5 2.6B is a 2.7B-parameter dense model from Liquid AI (United States), released 2026-07-28 under the LFM Open License license. It scores 43.2 on the Small LM Index, #12 of 21 tiny models. It needs about 2.3 GB of memory at 4-bit with an 8K-token context and costs $0.39 per million tokens through the cheapest API provider.

Small LM Index

43.2

#12 of 21 scored tiny models

Likely 39.0–47.5 · from 5 of 6 core benchmarks

Parameters

2.7B

Dense

Price per 1M tokens

$0.39

$0.35 in / $0.50 out · blended 3:1

Context window

128K

tokens

Benchmarks

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

GPQA Diamond Lab-reported Knowledge · From the LFM3 card57.8

Class best: Nanbeige4.2 3B 87.4

Humanity's Last Exam Lab-reported Reasoning · From the LFM3 card5.9

Class best: Spark-X2.6 4B 17.9

AIME Lab-reported Math · AIME 202551.9

Class best: Spark-X2.6 4B 91.2

LiveCodeBench Lab-reported Coding · LiveCodeBench v659.4

Class best: LFM3 1.7B 73.5

SWE-bench Verified Lab-reported Coding · OpenCode harness, from the LFM3 card18.0

Class best: Nanbeige4.2 3B 63.6

IFBench Lab-reported Instruction following59.2

Class best: Spark-X2.6 4B 75.8

IFEval Lab-reported Instruction following · From the LFM3 card95.4

Class best: LFM3 1.7B 96.7

BFCL Lab-reported Agentic · BFCL v456.9

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.

MMLU-Pro≈ 68

Specifications

Lab: Liquid AI (US · United States)
Architecture: Dense transformer
Total parameters: 2.7B
Active parameters: All (dense)
Size class: Tiny
Context window: 128K tokens
Input modalities: text
Reasoning: Instruct (no thinking)
License: LFM Open License
Openness: Open weights
Memory to run: ~2.3 GB at 4-bit, 8K context (more sizes)
Hugging Face: 69K downloads (30 days) · 859 likes
API providers: 1 on OpenRouter
API price: $0.35 in / $0.50 out per 1M (OpenRouter, 2026-11-03)

Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

LFM2.5 2.6B vs MiniCPM5 2B (2.5B, index 57.5)
LFM2.5 2.6B vs Granite 4.2 3B (3B, index 54.8)
LFM2.5 2.6B vs Ministral 3 3B (3B, index 49.4)
LFM2.5 2.6B vs SmolLM3 3B (3.1B, index 30.4)
LFM2.5 2.6B vs Qwen3.5 2B (2.3B, index 33.0)

Sources

https://huggingface.co/LiquidAI/LFM2.5-2.6B

Something wrong here? Report it.

Run it locally

Uses LiquidAI/LFM2.5-2.6B-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/LFM2.5-2.6B-GGUF:Q4_K_M

llama.cpp

llama-server -hf LiquidAI/LFM2.5-2.6B-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/LFM2.5-2.6B-MLX-4bit

vLLM (GPU server)

vllm serve LiquidAI/LFM2.5-2.6B

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	6.0 GB	6.5 GB	8.1 GB
8-bit Q8_0	3.5 GB	4.0 GB	5.5 GB
4-bit Q4_K_M	2.3 GB	2.7 GB	4.3 GB

Check it against your GPU, Mac or phone →