jina-vlm: Small Multilingual Vision Language Model

TL;DR

jina-vlm is a 2.4B parameter multilingual vision-language model that achieves state-of-the-art performance in VQA tasks among open models of similar scale, using efficient image processing techniques.

Contribution

The paper introduces jina-vlm, a token-efficient multilingual VLM with novel image tiling and attention pooling, and provides insights into data category importance through ablation studies.

Findings

Achieves state-of-the-art multilingual VQA performance among open 2B-scale VLMs.

Demonstrates effective token-efficient processing of arbitrary-resolution images.

Provides a systematic analysis of data category contributions via ablation studies.

Abstract

We present jina-vlm, a token-efficient 2.4B parameter vision-language model that achieves state-of-the-art multilingual VQA performance among open 2B-scale VLMs. The model couples a SigLIP2 vision encoder with a Qwen3 language decoder and makes use of image tiling and attention-pooling for token-efficient processing of arbitrary-resolution images. To understand the contribution of different training data categories, we conduct a leave-one-out data mixture ablation study-systematically removing task, domain, modality, and language categories-to diagnose which data types are necessary versus redundant and whether task benefits transfer across domains. Model weights and code are publicly released at https://huggingface.co/jinaai/jina-vlm.