Liquid AILFM2.5
LFM2.5-VL-3B
LFM2.5-VL-3B combines the LFM2.5-2.6B language backbone with a SigLIP2 NaFlex vision encoder.
What the release provides
It accepts text and images and supports OCR with layout annotations and object grounding. The published context is 32,768 tokens.LFM2.5-VL-3B official model card
How to use it well
Practical guidance. Review the annotation format before building a document parser around it. Validate extracted text and positions against the image. Keep image dimensions and coordinate conversion explicit so a correct detection does not become a misplaced application action.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | LiquidAI/LFM2.5-VL-3B |
| Context window | 32,768 tokens Downloadable checkpoint limit; the hosted API has a separate service contract. LFM2.5-VL-3B official model card |