[whatmodel.sh](../../README.md) / LFM2 8B-A1B # LFM2 8B-A1B Liquid AI · released 2025-10 · 8.3B params (1.5B active) · lfm1.0 **score 46.5** ±0.8 · comparison only: no current run option verified · from 31 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 8B-A1B among every model on that task or test, worst to best. 
task            vs the others          rank  counts  tests
agent tasks     ▁▁●▆█▄▅▄▄▃▁▁    #342 of 354    3.8%      4
chat & writing  ▁▁▁▃▅▄●█▇▇▇▃   #291 of 474*    4.9%      2
math            ▁▁▃▆█●▅█▃▁▁▁    #296 of 530     30%      7
agent tasks · #342 of 354 · 3.8% of its score Trails [Granite 4.0 350M](granite-4-0-350m.md), [Granite 4.0 H 350M](granite-4-0-h-350m.md), [ChatGLM3-6B](chatglm3-6b.md) here, though they rank lower overall. 
test                                result  vs the others in its run   difficulty        counts  why
τ²-bench Retail (maker)               7.0%  ●····▂▁▃▆█▃▁ beats 0/44    hard · avg 29%      1.2%  small run
BFCL V4 (December 2025 archive) (…   25.5%  ▁▁▁▂●▆▅▆█▂▄▃ beats 17/114  medium · avg 38%    1.1%  at its level · small run
BFCL v3 (maker)                      45.1%  ▁▁▁▁▁●▅█▆▅▄▁ beats 25/171  medium · avg 57%    0.7%  blunt · at its level · small run
τ²-bench Telecom (maker)             13.6%  ●█▄········█ beats 0/5     hard · avg 25%      0.7%  sharp · noisy
chat & writing · #291 of 474 · 4.9% of its score Beats [Mistral Small 4 (119B-A6.5B)](mistral-small-4.md), [Qwen2.5 Coder 32B Instruct](qwen2-5-coder-32b.md), [Ministral 3 14B Reasoning 2512](ministral-3-14b-reasoning.md) here, though they rank higher overall. Trails [Flan-T5 XXL](flan-t5-xxl.md), [Granite 4.0 H 1B](granite-4-0-h-1b.md), [LFM2.5 1.2B Instruct](lfm2-5-1-2b-instruct.md) here, though they rank lower overall. 
test                                result  vs the others in its run   difficulty      counts  why
IFEval mean of four measures (mak…   77.6%  █··▄●·▄····▄ beats 3/5     easy · avg 84%    3.7%  typical
IFEval (aggregation unspecified) …   79.4%  ▁▂▄▂▂▄●██▅▂▁ beats 50/145  easy · avg 82%    1.2%  small run
math · #296 of 530 · 30% of its score Beats [GPT-4o (2024-08-06)](gpt-4o-2024-08-06.md), [Command A (03-2025)](command-a.md), [GPT-4o (2024-05-13)](gpt-4o-2024-05-13.md) here, though they rank higher overall. Trails [Gemma 7B (base)](gemma-7b-base.md), [CodeGemma 7B (v1.0)](codegemma-7b.md), [Phi-2 (2.7B)](phi-2.md) here, though they rank lower overall. 
test                                result  vs the others in its run   difficulty           counts  why
GSM8K                                84.6%  █····●··▄·▄█ beats 3/7     very easy · avg 86%     11%  sharp
MATH (subset unspecified)            82.6%  █▄··●█▄····▄ beats 3/7     easy · avg 80%         5.1%  at its level
MATH-500 (maker)                     74.2%  ▃····▃·●···▃ beats 4/5     easy · avg 79%         4.4%  at its level · several runs
AIME 2025 (maker)                    20.0%  ▁·▁▁▁●▅▆█▇▅▁ beats 11/133  hard · avg 34%         4.0%  sharp · small run
MGSM (Marco evaluation)              56.5%  ▄·●·▄▄▄·▄··█ beats 1/7     easy · avg 69%         2.8%  at its level
PolyMath (Marco evaluation)          26.5%  ▄··▄●█··▄··▄ beats 3/7     hard · avg 29%         2.4%  typical
GSM8K loose substring accuracy (m…   84.4%  █·█··██●···█ beats 4/5     easy · avg 83%         0.4%  blunt
other tests (13) 
test                                result  vs the others in its run  difficulty        counts  why
MMLU                                 72.1%  ▄···▄●·▄█··█ beats 2/7    easy · avg 74%       15%  at its level · several runs
GPQA Diamond                         31.9%  ▄···●·▄·█·▄▄ beats 2/7    hard · avg 37%       11%  sharp · several runs
MMLU-Pro                             49.5%  ▄····●·▄█·▄▄ beats 1/7    medium · avg 54%    9.4%  at its level · several runs
IFBench (aggregation unspecified)…   25.9%  █··███··●··█ beats 4/5    hard · avg 31%      4.6%  at its level · several runs
MMLU-ProX-Lite (Marco)               33.8%  █··●··█··█·▄ beats 2/7    medium · avg 42%    4.0%  at its level
GlobalMMLU (Marco evaluation)        49.0%  ▃··▃●··▅···█ beats 2/7    medium · avg 56%    2.9%  at its level
Multi-IF three-turn average (make…   58.2%  ▄··●▄··▄···▄ beats 1/5    easy · avg 70%      2.8%  at its level
MMMLU seven languages (Liquid) (m…   55.3%  ▄·█····●···▄ beats 3/5    medium · avg 59%    2.7%  at its level
MMMLU (Marco evaluation)             56.5%  ▄···▄·●·▄·▄█ beats 3/7    medium · avg 59%    2.6%  at its level
AGIEval (Marco suite unspecified)    45.2%  ▄··▄█●··▄··█ beats 4/7    medium · avg 46%    2.1%  blunt · at its level
MGSM six languages, loose (Liquid…   72.4%  ▄·▄·●···▄··▄ beats 2/5    easy · avg 81%      1.8%  typical
Multi-IF (aggregation unspecified…   58.5%  ▁▁▁▁●▁▃▄█▄▁▂ beats 5/41   easy · avg 66%      1.0%  at its level · small run
GSM-Plus (maker)                     64.8%  █·····███·●█ beats 4/5    easy · avg 71%      0.6%  blunt · at its level
## 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
Marco-MoE: Open Multilingual Mixture-of…       11   54% ▇▇▇▇▇▇▇
LFM2 Technical Report (submitted / make…       11   32% ▇▇▇▇
LiquidAI/LFM2.5-8B-A1B model card (subm…        9   13% ▇▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 46.5
Licences: Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling ([CC BY 4.0](../../licenses/cards-final-NOTICE.txt)) · LFM2 Technical Report ([CC BY 4.0](../../licenses/cards-gap-NOTICE.txt)) · LiquidAI/LFM2.5-8B-A1B model card ([LFM Open License v1.0](../../licenses/cards-gap-NOTICE.txt)) ## How to run it No verified API or local download yet. [interactive version](https://whatmodel.sh/model/lfm2-8b-a1b) · [JSON](https://whatmodel.sh/api?format=json&all=1)