whatmodel.sh / LFM3 VL 2B
LFM3 VL 2B
↗ share
+ contribute
LFM3 VL 2B is an open-weights language model from Liquid AI, released October 2026. Its published size is 1.92B parameters and a 128k-token context window. Licence: lfm1.0.
Catalogue identity notes
Post-trained vision-language checkpoint. QAD Q4_0 edition is additionally trained and not conflated with original BF16 measurements; GPU serving support still in progress per card. Verified Ollama image-input route requires Ollama >=0.40.12 on macOS.
Model information sources
$ how it compares
Scored LLMs released closest to LFM3 VL 2B, out of 1192 with a score, then more from Liquid AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 3 |  GPT-6.1 Terra | 84±1 | api $2.00/M | ||
| 4 |  Gemini 4 Carbon | 83±1 | limited preview | ||
| 7 |  Muse Wonder Full | 82±2 | Muse Code only | ||
| 14 |  Kimi K3.1 Pro | 80±2 | api price unknown | ||
| 18 | Muse Wonder | 78±1 | api $8.00/M | ||
| more from Liquid AI | |||||
| 528 | LFM2.5 2.6B | 53±1 | local | ||
| 539 | LFM2.5 8B-A1B | 52±1 | local | ||
| 570 | LFM2 24B-A2B | 51±1 | local | ||
$ what it's good at
mathnot ranked on this task yet · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MathVista (testmini) lab | 84.2% ✎ | ▁··▁▄▄▆▇▄▅█▄ beats 37/48 | — |
vision & documentsnot ranked on this task yet · 7 tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CharXiv (reasoning questions) lab | 70.2% ✎ | ▂▅▃▅▅▆▆▅▆█▆▂ beats 57/101 | — |
| CharXiv descriptive questions lab | 87.5% ✎ | ▂▄▄▅▇▄██▇▂▁▂ beats 40/59 | — |
| MMMU (val) lab | 55.7% ✎ | ▁····▁▁▃█▆▆▅ beats 21/53 | — |
| MathVista (testmini) lab | 84.2% ✎ | ▁··▁▄▄▆▇▄▅█▄ beats 37/48 | — |
| RealWorldQA lab | 79.3% ✎ | ▁▃▁▇▃▇▇▅█▁▅▅ beats 28/39 | — |
| MMStar lab | 74.4% ✎ | ▃▅▂▃▂▂▆██▆█▅ beats 19/35 | — |
| HallusionBench lab | 65.6% ✎ | ▂▁·▂▂▂▃▅▃█▅▅ beats 23/33 | — |
$ where to run it
No verified API or download recorded yet · add one
Official weights and documentation: huggingface.co/LiquidAI ↗
$ published results · 7 results, not enough overlap for a score
| test | result | source and conditions |
|---|---|---|
| MMStar | 74.4% | Liquid AI own-model benchmark facts · lab-reported Liquid own thinking-enabled LFM3 VL 2B column, full 62-benchmark table, not category means. MMMU val, MathVista mini, CharXiv val. Published model-specific preprocessing; GPT-5.6 Luna judge for judge-based tests. No competitor columns. |
| MMMU (val) | 55.7% | Liquid AI own-model benchmark facts · lab-reported Liquid own thinking-enabled LFM3 VL 2B column, full 62-benchmark table, not category means. MMMU val, MathVista mini, CharXiv val. Published model-specific preprocessing; GPT-5.6 Luna judge for judge-based tests. No competitor columns. |
| MathVista (testmini) | 84.2% | Liquid AI own-model benchmark facts · lab-reported Liquid own thinking-enabled LFM3 VL 2B column, full 62-benchmark table, not category means. MMMU val, MathVista mini, CharXiv val. Published model-specific preprocessing; GPT-5.6 Luna judge for judge-based tests. No competitor columns. |
| RealWorldQA | 79.3% | Liquid AI own-model benchmark facts · lab-reported Liquid own thinking-enabled LFM3 VL 2B column, full 62-benchmark table, not category means. MMMU val, MathVista mini, CharXiv val. Published model-specific preprocessing; GPT-5.6 Luna judge for judge-based tests. No competitor columns. |
| HallusionBench | 65.6% | Liquid AI own-model benchmark facts · lab-reported Liquid own thinking-enabled LFM3 VL 2B column, full 62-benchmark table, not category means. MMMU val, MathVista mini, CharXiv val. Published model-specific preprocessing; GPT-5.6 Luna judge for judge-based tests. No competitor columns. |
| CharXiv (reasoning questions) | 70.2% | Liquid AI own-model benchmark facts · lab-reported Liquid own thinking-enabled LFM3 VL 2B column, full 62-benchmark table, not category means. MMMU val, MathVista mini, CharXiv val. Published model-specific preprocessing; GPT-5.6 Luna judge for judge-based tests. No competitor columns. |
| CharXiv descriptive questions | 87.5% | Liquid AI own-model benchmark facts · lab-reported Liquid own thinking-enabled LFM3 VL 2B column, full 62-benchmark table, not category means. MMMU val, MathVista mini, CharXiv val. Published model-specific preprocessing; GPT-5.6 Luna judge for judge-based tests. No competitor columns. |
$ 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