AI BriefingTencent Releases High-Performance VDR Model EVIE
·2026.08.17 21:24
Key point
Tencent has released an ultra-lightweight, high-performance multilingual Visual Document Retrieval (VDR) model based on Qwen3.5-4B.
Details
EVIE-Preview-4.5B is a state-of-the-art multilingual Visual Document Retrieval (VDR) model built on Qwen3.5-4B.
It adopts a ColBERT-style late interaction approach to generate ultra-compact 128-dimensional (128D) multi-vector token embeddings. This reduces vector storage and indexing costs by 8 to 32 times compared to existing 2560D–4096D methods.
Key Features:
- SOTA Performance: Outperformed existing 8B models on 7 out of 8 public domains in the ViDoRe V3 benchmark.
- Multilingual and Multi-format Support: Supports various languages such as English, Chinese, and German, as well as diverse visual formats including charts, tables, scientific reports, and financial documents.
- ColPali Compatibility: Fully integrates with the standard
colpali-engineecosystem and late interaction scoring pipelines.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Apple Releases LensVLM-9B for Processing Compressed Text Images
Qwen Releases 'RecreationWorld' for Training AI Agents to Rebuild AppsPrivatemode Benchmarks GLM-5.3-Flash as Vision-Capable Decision Model with Parity to Jev but Higher Cost
Xiaomi Releases Open-Weight Multimodal Model MiMo-V2.6-Pro… Higher Intelligence and Competitive Pricing Compared to Similar-Sized Models