WeVisOne-VL - Segmentation RefCOCO testA: leaderboard
WeVisOne-VL official release:greedy decoding,temperature0,max1960visual tokens/image;Qwen3-VL Instruct and Qwen3.5 Thinking baselines. Generative visual grounding/perception,not image generation. Classical detectors and specialist OCR systems are outside the VLM cohort;release checkpoints only,reported comparator scores not copied. RefCOCO andRef-ZOM cIoU,ReasonSeg gIoU,absent-query rejection separately. Score (%).
Metric: Score (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | WeVisOne-VL-4B | 84.5 | |
| 2 | WeVisOne-VL-2B | 82.9 |
Interactive version: theaggregate.ai/benchmark?slug=wevisone-vl-segmentation-refcoco-testa · How It Works · Data refreshed daily, snapshot 2026-10-24.