Tencent EVIE-Preview-4.5B tops ViDoRe V3 with compact 128-dim multilingual visual document retrieval
Tencent
EVIE-Preview-4.5B is a multilingual visual document retrieval model built on Qwen3.5-4B (ColBERT-style late interaction, interleaved Gated DeltaNet with full attention) producing 128-dim multi-vector token embeddings. It scores 64.40 nDCG@10 average on ViDoRe V3 (leading 7 of 8 domains) and 85.93 average on ViDoRe V1 plus V2, covering EN, FR, DE, IT, ES, PT, ZH.
Why it matters
Compact 4.5B VDR model that beats 8B-class competitors while staying compatible with the existing colpali-engine pipeline, so it slots into production retrieval stacks with minimal infra changes; the preview status implies a v1 to follow.
Importance: 3/5
open-weights / GA marker
Sources
official
tencent/EVIE-Preview-4.5B