whatmodel.sh / LFM2 8B-A1B
LFM2 8B-A1B
↗ share
+ contribute
LFM2 8B-A1B is an open-weights language model from Liquid AI, released October 2025. Its published size is 8.3B parameters (1.5B active per token). Licence: lfm1.0.
Catalogue identity notes
Comparison-only, proposed so the card-liquidai-lfm2-5-8b-a1b runs count. The pinned LFM2.5-8B-A1B card (@5dd2260) prints 'Improvements over LFM2-8B-A1B' with own LFM2-8B-A1B and LFM2.5-8B-A1B columns. Weights: https://huggingface.co/LiquidAI/LFM2-8B-A1B (sha c1c44ff9fc00db3ebf4516970563f5f383d23670, created 2025-10-07, licence 'other' = LFM Open License v1.0, 8,339,929,856 params in safetensors metadata; 8.3B total / 1.5B active per Liquid's naming).
$ how it compares
The LLMs ranked either side of LFM2 8B-A1B, out of 828 with a score, then more from Liquid AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 515 |  Granite 3.3 8B Instruct | 46±1 | local | ||
| 516 |  Tulu 2 70B | 46±2 | local | ||
| 517 | LFM2 8B-A1B | 46±1 | comparison only | ||
| 518 |  OpenChat 3.5 0106 Gemma | 46±3 | local | ||
| 519 | Olmo 3 7B Instruct | 46±1 | local | ||
| more from Liquid AI | |||||
| 370 | LFM2.5 2.6B | 53±1 | local | ||
| 375 | LFM2.5 8B-A1B | 52±1 | local | ||
| 398 | LFM2 24B-A2B | 51±1 | local | ||
$ what it's good at
agent tasks▁▁▃▆█▆▅▄▄▃▂▁ #328 of 340 · 3.8% of its score · 4 tests
Trails Granite 4.0 350M, Granite 4.0 H 350M, ChatGLM3-6B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| τ²-bench Retail lab | 7.0% ✎ | ▃····▂▁▃▆█▃▁ beats 0/44 | 1.3% |
| BFCL V4 (December 2025 archive) lab | 25.5% ✎ | ▁▁▁▂▇▆▆▆█▂▄▂ beats 17/111 | 1.2% |
| BFCL v3 lab | 45.1% ✎ | ▁▁▁▁▁▃▅█▆▅▄▁ beats 25/170 | 0.7% |
| τ²-bench Telecom lab | 13.6% ✎ | ▄█▄········█ beats 0/5 | 0.6% |
chat & writing▁▁▁▃▅▄▆█▇▇▇▃ #284 of 463 * · 4.8% of its score · 2 tests
Beats Mistral Small 4 (119B-A6.5B), Ministral 3 14B Reasoning 2512, Qwen2.5 Coder 32B Instruct here, though they rank higher overall. Trails Flan-T5 XXL, Granite 4.0 H 1B, LFM2.5 1.2B Instruct here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| IFEval mean of four measures lab | 77.6% ✎ | █··▄▄·▄····▄ beats 3/5 | 3.5% |
| IFEval (aggregation unspecified) lab | 79.4% ✎ | ▁▁▃▁▂▄▄▆█▃▂▁ beats 50/141 | 1.3% |
math▁▂▃▅▇▅▅█▄▁▁▁ #277 of 462 · 31% of its score · 7 tests
Beats Command A (03-2025), GPT-4o (2024-11-20), GPT-4o (2024-05-13) here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Tulu 2 DPO 70B, OpenThinker 7B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GSM8K | 84.6% ✎ | █····█··▄·▄█ beats 3/7 | 9.5% |
| MATH-500 lab +1 | 74.2% ✎ | ▃····▃·█···▃ beats 4/5 | 5.9% |
| MATH (subset unspecified) | 82.6% ✎ | █▄··▄█▄····▄ beats 3/7 | 5.2% |
| AIME 2025 lab | 20.0% ✎ | ▁·▁▁▁▃▅▆█▇▅▁ beats 11/133 | 4.2% |
| MGSM (Marco evaluation) | 56.5% ✎ | ▄·▄·▄▄▄·▄··█ beats 1/7 | 3.1% |
| PolyMath (Marco evaluation) | 26.5% ✎ | ▄··▄██··▄··▄ beats 3/7 | 2.6% |
| GSM8K loose substring accuracy lab | 84.4% ✎ | █·█··███···█ beats 4/5 | 0.3% |
other tests13 not in a task · 60% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MMLU +1 | 72.1% ✎ | ▄···▄▄·▄█··█ beats 2/7 | 14% |
| GPQA Diamond +1 | 31.9% ✎ | ▄···█·▄·█·▄▄ beats 2/7 | 11% |
| MMLU-Pro +1 | 49.5% ✎ | ▄····█·▄█·▄▄ beats 1/7 | 9.5% |
| IFBench (aggregation unspecified) lab +1 | 25.9% ✎ | █··███··█··█ beats 4/5 | 4.7% |
| MMLU-ProX-Lite (Marco) | 33.8% ✎ | █··▄··█··█·▄ beats 2/7 | 4.3% |
| GlobalMMLU (Marco evaluation) | 49.0% ✎ | ▃··▃▃··▅···█ beats 2/7 | 3.1% |
| Multi-IF three-turn average lab | 58.2% ✎ | ▄··█▄··▄···▄ beats 1/5 | 3.0% |
| MMMLU seven languages (Liquid) lab | 55.3% ✎ | ▄·█····█···▄ beats 3/5 | 2.7% |
5 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MMMLU (Marco evaluation) | 56.5% ✎ | ▄···▄·█·▄·▄█ beats 3/7 | 2.7% |
| AGIEval (Marco suite unspecified) | 45.2% ✎ | ▄··▄█▄··▄··█ beats 4/7 | 2.2% |
| MGSM six languages, loose (Liquid) lab | 72.4% ✎ | ▄·▄·█···▄··▄ beats 2/5 | 1.9% |
| Multi-IF (aggregation unspecified) lab | 58.5% ✎ | ▁▁▁▁▃▁▃▄█▄▁▂ beats 5/41 | 1.1% |
| GSM-Plus lab | 64.8% ✎ | █·····███·██ beats 4/5 | 0.6% |
$ where to run it
No verified API or download recorded yet · add one
$ how the score was worked out · Marco-MoE: Open Multilingual… 53%, LFM2 Technical Report — Liqui… 32%, LiquidAI/LFM2.5-8B-A1B model… 14%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The 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's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling | 11 | 53% |
| LFM2 Technical Report — Liquid AI own SmolLM3 evaluation · lab-reported | 11 | 32% |
 LiquidAI/LFM2.5-8B-A1B model card · lab-reported | 9 | 14% |
| prior · pull toward the middle | 0% | |
| weighted average | = 45.6 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json