[whatmodel.sh](../../README.md) / LFM2 350M
# LFM2 350M
Liquid AI · released 2025-07 · 0.354B params · lfm1.0
**score 32.8** ±1.5 · #645 of 732 · from 10 benchmark results across 3 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks LFM2 350M among every model on that task or test, worst to best.
task vs the others rank counts tests
chat & writing ▁▁▁▃▆●▆█▆▇▇▃ #302 of 402* 8.1% 1
math ▁▁▃●█▆▄█▄▂▁▁ #378 of 431* 43% 2
chat & writing · #302 of 402 · 8.1% of its score
Beats [Ministral 3 14B Reasoning 2512](ministral-3-14b-reasoning.md), [Gemma 4 E4B](gemma-4-e4b.md), [OpenThinker3 7B](openthinker3-7b.md) here, though they rank higher overall.
test result vs the others in its run difficulty counts why
IFEval (aggregation unspecified) … 65.1% ▄····█●···▄█ beats 3/6 easy · avg 82% 8.1% at its level
math · #378 of 431 · 43% of its score
Beats [Command R Plus (04-2024)](command-r-plus-04-2024.md), [Yi 34B Chat](yi-34b-chat.md), [Llama 3 8B Instruct](llama-3-8b.md) here, though they rank higher overall. Trails [GPT-Neo 1.3B](gpt-neo-1-3b.md), [OPT-1.3B](opt-1-3b.md), [OPT-2.7B](opt-2-7b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
MGSM (maker) 29.5% ●···█▄··▄··▄ beats 1/6 easy · avg 83% 32% sharp · at its level
GSM8K (maker) 30.1% ●·█···▄·▄··█ beats 0/6 easy · avg 83% 11% sharp · at its level
other tests (7)
test result vs the others in its run difficulty counts why
MMLU (maker) 43.4% █·●██·█··█·█ beats 1/6 easy · avg 74% 19% typical
IFBench (aggregation unspecified)… 16.4% ●▄·▄····▄·█▄ beats 0/6 medium · avg 38% 8.0% typical
MMMLU (maker) 38.0% ▄···▄·●▄··▄█ beats 2/6 medium · avg 57% 7.7% typical
Open PT-LLM Leaderboard average (… 32.2% ●▁▁▁▁▁▁▁▂█▅▁ beats 18/276 medium · avg 72% 3.9% noisy
GPQA (subset unspecified) (maker) 27.5% █······●▄█·▄ beats 2/6 hard · avg 37% 3.6% sharp · too hard for it · noisy
EuroEval Life in the UK (English … 37.8% ▁▁▁▁▁●▁▂▄█▄▃ beats 13/146 easy · avg 82% 3.5% at its level · noisy
EuroEval HellaSwag (English), tes… 25.6% ●·▁▁▂▁▂▅█▇▄▁ beats 7/100 medium · avg 67% 2.2% sharp · noisy
source results share of the score
LiquidAI/LFM2-1.2B model card (submitte… 7 89% ▇▇▇▇▇▇▇▇▇▇▇
EuroEval leaderboards 2 6% ▇
Open Portuguese LLM Leaderboard 1 4% ▇
prior: pull toward the middle 1% ▇
weighted average = 32.8
Licences: LiquidAI/LFM2-1.2B model card ([LFM Open License v1.0](https://huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/LICENSE)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE)) · Open Portuguese LLM Leaderboard ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt))
## How to run it
### llama.cpp (recommended)
1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md)
```sh
# macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
# Windows
irm https://llama.app/install.ps1 | iex
```
2. Run LFM2 350M (Q4_K_M, needs ~0.23 GB, from [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2-350M-GGUF/tree/8fdc9d526b7ed346b19257551b05816c7912ecc2))
```sh
llama cli -hf LiquidAI/LFM2-350M-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf LiquidAI/LFM2-350M-GGUF:Q4_K_M
```
### LM Studio
1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download)
2. Run LFM2 350M (Q4_K_M, needs ~0.23 GB, from [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2-350M-GGUF/tree/8fdc9d526b7ed346b19257551b05816c7912ecc2))
```sh
lms get https://huggingface.co/LiquidAI/LFM2-350M-GGUF@Q4_K_M -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: 4bit (~0.2 GB, Mac only) `lms get https://huggingface.co/mlx-community/LFM2-350M-4bit -y` · 8bit (~0.4 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/LFM2-350M-MLX-8bit -y`
### Ollama
1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download)
```sh
# Linux
curl -fsSL https://ollama.com/install.sh | sh
```
2. Run LFM2 350M (Q4_K_M, needs ~0.23 GB, from [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2-350M-GGUF/tree/8fdc9d526b7ed346b19257551b05816c7912ecc2))
```sh
ollama run hf.co/LiquidAI/LFM2-350M-GGUF:Q4_K_M
```
### MLX (Mac)
1. Install MLX (Mac). Apple silicon Macs only, text only.
```sh
# with uv
uv tool install mlx-lm
# or
pip install mlx-lm
```
2. Run LFM2 350M (4bit, needs ~0.2 GB, Mac only, from [huggingface.co/mlx-community](https://huggingface.co/mlx-community/LFM2-350M-4bit/tree/18dc72abf3b2337f9123cfd6eeeb58dfa7947066))
```sh
mlx_lm.chat --model mlx-community/LFM2-350M-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model mlx-community/LFM2-350M-4bit
```
Other versions: 8bit (~0.4 GB, Mac only) `mlx_lm.chat --model lmstudio-community/LFM2-350M-MLX-8bit`
Model repository and documentation: [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2-350M)
[interactive version](https://whatmodel.sh/model/lfm2-350m) · [JSON](https://whatmodel.sh/api?format=json&all=1)