LFM2.5-VL-1.6B: benchmark results
Liquid AI's LFM2.5-VL-1.6B, an updated vision-language model built on the LFM2.5-1.2B-Base backbone with a SigLIP2 NaFlex vision encoder, supporting 8 languages and a 32,768-token context. Provider: Liquid AI. Released 2026-01-05. Access: Open.
Aggregate Rating 1478 ± 25, rank #854 of 1656 rated models, from 65 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| TomaBench | 61.3 | Accuracy (%; zero-shot multiple-choice questions on real tom | 87.5 |
| TomaBench - Crop Species Identification | 76.5 | Accuracy (%; zero-shot multiple-choice questions on real tom | 87.5 |
| TomaBench - Disease Classification | 41.9 | Accuracy (%; zero-shot multiple-choice questions on real tom | 75 |
| TomaBench - Pathogen Classification | 54.1 | Accuracy (%; zero-shot multiple-choice questions on real tom | 75 |
| TomaBench - Scientific Name Classification | 38.1 | Accuracy (%; zero-shot multiple-choice questions on real tom | 50 |
| TomaBench - Symptom Identification | 33 | Accuracy (%; zero-shot multiple-choice questions on real tom | 50 |
| AI for Education Visual Reasoning - match (process) | 30.6 | Accuracy (%) | 34.6 |
| AA Humanity's Last Exam | 5.1 | Accuracy (%) | 28.9 |
| AI for Education Visual Reasoning - pattern completion (2d) | 31.5 | Accuracy (%) | 26.2 |
| AA CritPt | 0 | Accuracy (%) | 22 |
| AA IFBench | 33.13 | Accuracy (%) | 21.7 |
| Roboflow Vision Evals - Visual Understanding | 53.73 | Pass Rate (%) | 21.1 |
Interactive version: theaggregate.ai/model?slug=lfm2-5-vl-1-6b · How It Works · Data refreshed daily, snapshot 2026-10-20.