[whatmodel.sh](../../README.md) / LFM2.5 8B-A1B # LFM2.5 8B-A1B Liquid AI · released 2026-05 · 8.47B params (1.5B active) · lfm1.0 **score 52.3** ±0.9 · #333 of 732 · from 29 benchmark results across 2 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks LFM2.5 8B-A1B among every model on that task or test, worst to best. 
task            vs the others          rank  counts  tests
agent tasks     ▁▁▃▆●▄▅▅▄▄▂▁    #230 of 313     17%      8
coding          ▁▁▁▃●▇█▇▅▄▃▁    #393 of 469     18%      5
chat & writing  ▁▁▁▃▆▄▆█▆●▇▃   #132 of 402*    1.7%      1
math            ▁▁▃▅█●▄█▄▂▁▁    #223 of 431     42%      4
web search      ▅·▅●█·▃▃▃·▅▃     #11 of 15*    2.3%      1
agent tasks · #230 of 313 · 17% of its score Beats [o1 (2024-12-17)](o1-2024-12-17.md), [o3-mini](o3-mini.md), [DeepSeek R1 (Jan '25)](deepseek-r1.md) here, though they rank higher overall. Trails [SmolLM3 3B](smollm3-3b.md), [Jamba 1.5 Mini](jamba-1-5-mini.md), [LFM2.5 1.2B Thinking](lfm2-5-1-2b-thinking.md) here, though they rank lower overall. 
test                               result  vs the others in its run    difficulty            counts  why
BFCL V4 (December 2025 archive)     49.2%  ▄·█·▄●▄▄·▄·▄ beats 4/8      medium · avg 43%        5.2%  at its level · several runs
SWE-bench Verified                   0.4%  ●··▄▄█··▄·▄█ beats 0/8      very hard · avg 11%     2.6%  sharp · noisy
BrowseComp-ZH                       13.2%  ▄▄·▄▄●·█···█ beats 4/8      very hard · avg 13%     2.3%  noisy
τ²-bench Retail (maker)             39.8%  ▁····●▁▃▆█▃▁ beats 3/44     hard · avg 29%          1.8%  at its level · small run
SWE-bench Pro (split unspecified)    0.4%  ●▂···▂···▄·▂ beats 0/8      very hard · avg 3.9%    1.7%  sharp · too hard for it · noisy
τ³-bench Banking                     3.4%  ▂▂·●█▂▃▃▃▆▂▅ beats 2/24     very hard · avg 3.6%    1.4%  blunt · too hard for it
τ²-bench Telecom (maker)            88.1%  ▄█▄········● beats 5/5      hard · avg 25%          1.4%  sharp · at its level · noisy
BFCL v3 (maker)                     64.4%  ▁▁▁▁▁▃▅█●▅▄▁ beats 120/171  medium · avg 57%        0.9%  blunt · at its level · small run
coding · #393 of 469 · 18% of its score Trails [Qwen2.5 1.5B Instruct](qwen2-5-1-5b.md), [WizardLM 13B V1.2](wizardlm-13b-v1-2.md), [Falcon3 3B Instruct](falcon3-3b.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty            counts  why
LiveCodeBench v6 (window unspecif…   39.8%  ▃····●·▃▅▃·▃ beats 1/8    medium · avg 38%        6.7%  at its level
LiveCodeBench Pro 2025Q2 (Easy)      27.8%  ▄···●▄··▄█▄▄ beats 2/8    hard · avg 33%          4.6%  at its level
LiveCodeBench Pro 2025Q2 (Medium)     0.0%  ●···▂··▄▂··▂ beats 0/8    very hard · avg 2.0%    2.7%  sharp · too hard for it · noisy
SWE-bench Verified                    0.4%  ●··▄▄█··▄·▄█ beats 0/8    very hard · avg 11%     2.6%  sharp · noisy
SWE-bench Pro (split unspecified)     0.4%  ●▂···▂···▄·▂ beats 0/8    very hard · avg 3.9%    1.7%  sharp · too hard for it · noisy
chat & writing · #132 of 402 · 1.7% of its score Beats [LongCat 2.0 (1.6T-A48B)](longcat-2-0.md), [o3-mini](o3-mini.md), [K-EXAONE (236B-A23B)](k-exaone-236b-a23b.md) here, though they rank higher overall. 
test                              result  vs the others in its run  difficulty      counts  why
IFEval (aggregation unspecified)   90.8%  ▄▄····▄··█●█ beats 6/8    easy · avg 82%    1.7%  small run · several runs
math · #223 of 431 · 42% of its score Beats [MiniMax-M2.1](minimax-m2-1.md), [DeepSeek V3 (Dec '24)](deepseek-v3.md), [GPT-4o (2024-08-06)](gpt-4o-2024-08-06.md) here, though they rank higher overall. Trails [Phi-2 (2.7B)](phi-2.md), [Granite 8B Code Base (4K)](granite-8b-code-base-4k.md), [Mixtral 8x7B v0.1 (base)](mixtral-8x7b-v0-1.md) here, though they rank lower overall. 
test                result  vs the others in its run  difficulty           counts  why
AIME 2025            46.0%  █▄▄●▄····█·▄ beats 4/8    medium · avg 42%        15%  sharp · at its level · several runs
HMMT February 2026   38.5%  ▅··▃▃·●··▃·█ beats 4/8    hard · avg 30%          11%  sharp · at its level
AIME 2026            56.7%  ▄·▄█·●▄···█▄ beats 4/8    medium · avg 48%       8.4%  sharp · at its level
MATH-500             93.2%  ██▄●▄··▄···▄ beats 5/8    very easy · avg 89%    7.0%  several runs
web search · #11 of 15 · 2.3% of its score Beats [Nemotron 3 Nano 4B](nemotron-3-nano-4b.md), [Gemma 4 E4B](gemma-4-e4b.md), [LFM2.5 2.6B](lfm2-5-2-6b.md) here, though they rank higher overall. Trails [Qwen3.5 2B](qwen3-5-2b.md) here, though they rank lower overall. 
test           result  vs the others in its run  difficulty           counts  why
BrowseComp-ZH   13.2%  ▄▄·▄▄●·█···█ beats 4/8    very hard · avg 13%    2.3%  noisy
other tests (7) 
test                                result  vs the others in its run  difficulty           counts  why
MMLU-Pro                             63.1%  ▂··●█·▂▂···▂ beats 1/8    medium · avg 59%       5.2%  at its level
MMLU-Redux                           80.0%  ▂···●··▂▂··▂ beats 3/8    easy · avg 76%         4.7%  at its level
IFBench (aggregation unspecified)    51.0%  ▅····▃●·█·▃▃ beats 3/8    medium · avg 44%       3.7%  at its level · several runs
OJBench                               8.2%  ▃····●▅··█▃▃ beats 1/8    very hard · avg 12%    3.2%  typical
SuperGPQA                            34.5%  ▃·▃·●▅█····▃ beats 2/8    hard · avg 32%         2.9%  at its level
Multi-IF (aggregation unspecified)   71.4%  ▄▄···▄·▄·●██ beats 4/8    medium · avg 60%       2.7%  at its level · several runs
LongBench v2                         30.4%  ▃······●▅▃·█ beats 2/8    very hard · avg 32%    2.5%  typical
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its *counts* column above; here it is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
openbmb/MiniCPM5-2B model card                 20   82% ▇▇▇▇▇▇▇▇▇▇
LiquidAI/LFM2.5-8B-A1B model card (subm…        9   18% ▇▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 52.3
Licences: openbmb/MiniCPM5-2B model card ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)) · LiquidAI/LFM2.5-8B-A1B model card ([LFM Open License v1.0](../../licenses/cards-gap-NOTICE.txt))
Not counted (1) 
source                  test       score  why
LiquidAI/LFM2.5-8B-A1…  AIME 2026  50.0%  only model in this source's run
## 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 LFM2.5 8B-A1B (Q4_K_M, needs ~5.2 GB, from [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-GGUF/tree/49c14831707011e64d70b2ebd8462ba08d608434)) ```sh llama cli -hf LiquidAI/LFM2.5-8B-A1B-GGUF:Q4_K_M ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf LiquidAI/LFM2.5-8B-A1B-GGUF:Q4_K_M ``` Other versions: Q8_0 (~9 GB) `llama cli -hf LiquidAI/LFM2.5-8B-A1B-GGUF:Q8_0` ### 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 LFM2.5 8B-A1B (Q4_K_M, needs ~5.2 GB, from [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-GGUF/tree/49c14831707011e64d70b2ebd8462ba08d608434)) ```sh lms get https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-GGUF@Q4_K_M -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: Q8_0 (~9 GB) `lms get https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-GGUF@Q8_0 -y` · 4bit (~4.8 GB, Mac only) `lms get https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-MLX-4bit -y` · 8bit (~9 GB, Mac only) `lms get https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-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 LFM2.5 8B-A1B (Q4_K_M, needs ~5.58 GB, from [ollama.com](https://ollama.com/library/lfm2.5/tags)) ```sh ollama run lfm2.5:8b ``` Other versions: Q8_0 (~10.66 GB) `ollama run lfm2.5:8b-a1b-q8_0` ### 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 LFM2.5 8B-A1B (4bit, needs ~4.8 GB, Mac only, from [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-MLX-4bit/tree/2e92b640a63d47ad4dcf81a19a366b902356b3bc)) ```sh mlx_lm.chat --model LiquidAI/LFM2.5-8B-A1B-MLX-4bit ``` 3. Or start an OpenAI-compatible server at http://localhost:8080/v1 ```sh mlx_lm.server --model LiquidAI/LFM2.5-8B-A1B-MLX-4bit ``` Other versions: 8bit (~9 GB, Mac only) `mlx_lm.chat --model LiquidAI/LFM2.5-8B-A1B-MLX-8bit` Model repository and documentation: [huggingface.co/LiquidAI](https://huggingface.co/LiquidAI/LFM2.5-8B-A1B) ## Published speed measurements Reference results for the hardware, engine and settings below. These are not verified speeds of the downloads above or predictions for your machine. 
hardware                      precision         speed  engine and settings
AMD Ryzen AI Max+ 395 (Stri…  Q4_K_M     171.17 tok/s  llama.cpp ac4cddeb0 (build_number…
AMD Ryzen AI Max+ 395 (Stri…  Q4_K_M     168.96 tok/s  llama.cpp 2016bf2 (build_number 1…
AMD Ryzen AI Max+ 395 (Stri…  Q4_K_M     176.48 tok/s  llama.cpp 98d5e8ba8 (build_number…
Apple M5 Max (40-core GPU),…  q4_k_m     250.12 tok/s  ollama (version not recorded) · l…
Each speed links to its source, which has the full settings. [interactive version](https://whatmodel.sh/model/lfm2-5-8b-a1b) · [JSON](https://whatmodel.sh/api?format=json&all=1)