Atlas

Models

← All models

LFM2.5-VL-1.6B

Liquid AI · 2026-01-05 · 1.6B parameters

LFM2.5-VL-1.6B is Liquid AI's 1.6-billion-parameter vision-language model from the LFM2.5 on-device family, designed for low-latency multimodal inference on edge devices. It extends the LFM2.5 hybrid text architecture with a visual encoder, enabling image understanding alongside language tasks while remaining deployable on CPUs and NPUs. The model was released alongside other LFM2.5 variants as part of Liquid AI's push to bring production-ready multimodal agents to on-device hardware.

Benchmark scores

BenchmarkScore
AA-LCR0.0
AA-Omniscience Index-83.9
Artificial Analysis Intelligence Index1.0
Artificial Analysis Omniscience Accuracy5.2
Artificial Analysis Omniscience Hallucination Rate94.0
Artificial Analysis Openness Index27.8
Capability106.1
CritPt0.0
GPQA Diamond28.9
Humanity's Last Exam (Text-Only)5.1
IFBench (Artificial Analysis)33.1
MMMU Pro26.5
SciCode (Artificial Analysis)3.0
Terminal-Bench Hard0.0
τ²-bench (Telecom)8.5
Loading Atlas data…