LFM2.5-VL-3B

Multimodal
Liquid AI

LFM2.5-VL-3B is Liquid AI's open-weight vision-language model (~3.12B parameters; HF safetensors total 3,123,483,888) for on-device and edge image-text→text. Non-reasoning: answers directly for low latency. Builds on the LFM2.5-2.6B language backbone with a SigLIP2 400M NaFlex vision encoder; improves screen/UI understanding, grounding, function calling, and multi-image input over LFM2-VL-3B. Pre-trained on ~34T tokens; 128K vocabulary; 32,768-token context. Day-one support for llama.cpp, MLX, vLLM, SGLang, and ONNX. License: LFM Open License v1.0 (lfm1.0).

Key Specifications

Parameters
3.1B
Context
-
Release Date
August 12, 2026
Average Score
63.7%

Timeline

Key dates in the model's history
Announcement
August 12, 2026
Last Update
September 10, 2026
Today
September 19, 2026

Technical Specifications

Parameters
3.1B
Training Tokens
34.0T tokens
Knowledge Cutoff
-
Family
-
Capabilities
MultimodalZeroEval

Benchmark Results

Model performance metrics across various tests and benchmarks

Multimodal

Working with images and visual data
ChartQA
Vendor table (ChartQA test)Self-reported
81.3%
DocVQA
Vendor table (DocVQA val); Liquid footnote: original 0–1 normalized to 0–100Self-reported
91.1%

Other Tests

Specialized benchmarks
BFCL-V4
Vendor table (BFCL V4)Self-reported
32.5%
BLINK
Vendor table (BLINK)Self-reported
61.5%
Hallusion Bench
Vendor table (HallusionBench)Self-reported
47.2%
IFBench
Vendor table (IFBench)Self-reported
25.8%
IFEval
Vendor table (IFEval)Self-reported
82.3%
MathVista-Mini
Vendor table (MathVista mini)Self-reported
68.5%
MMBench-V1.1
Vendor table (MMBench dev EN v1.1)Self-reported
81.0%
MME
Vendor table (MME); Liquid normalizes 0–2800 → 0–100Self-reported
73.1%
MM IF-Eval
Vendor table (MM-IFEval)Self-reported
60.6%
MMMU-Pro
Vendor table (MMMU-Pro); Liquid footnote: 10-choice MC variantSelf-reported
30.5%
MMMU (val)
Vendor table (MMMU val)Self-reported
48.4%
MMStar
Vendor table (MMStar)Self-reported
63.3%
MuirBench
Vendor table (MuirBench)Self-reported
58.3%
Multi-IF
Vendor table (Multi-IF)Self-reported
59.4%
OCRBench
Vendor table (OCRBench v1); Liquid normalizes 0–1000 → 0–100Self-reported
84.2%
OCRBench-V2 (en)
Vendor table (OCRBench v2 En)Self-reported
47.5%
POPE
Vendor table (POPE); Liquid footnote: F1 over adversarial/popular/randomSelf-reported
88.7%
RealWorldQA
Vendor table (RealWorldQA)Self-reported
73.1%
RefCOCO-avg
Vendor table (RefCOCO-avg P@1 over 8 splits)Self-reported
87.9%
SimpleVQA
Vendor table (SimpleVQA)Self-reported
35.4%
TextVQA
Vendor table (TextVQA val)Self-reported
84.3%

License & Metadata

License
lfm1_0
Announcement Date
August 12, 2026
Last Updated
September 10, 2026

Compare LFM2.5-VL-3B

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.