FindIt - Referring Expression Detection: leaderboard

Metric: Average F1@0.5 (%) over referring expression detection on 1,000-query subsets of RefCOCO/+/g (averaged), RefL4, Flickr30k-Entities, D3, PhraseCut and SVG; each model scored with the box representation, text or JSON output and JSON key that a two-stage probe on Pascal VOC found best for it; boxes Hungarian-matched to the ground truth, a match counts at IoU 0.5 or more (with the right label for multi-label queries); unparseable outputs count as empty; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 10 models tracked.

Top models

#ModelScore
1Qwen 3 VL 8B Instruct67.8
2Qwen 3.5 9B (Thinking)67
3Qwen 3.5 9B (Non-reasoning)64.5
4InternVL3-8B60.8
5Gemini 2.5 Flash (Non-reasoning)60.2
6Gemma 4 E4B59
7Qwen 2.5 VL 7B Instruct54.2
8GPT-5.4 (Non-reasoning)44.3
9Claude Sonnet 4.523.2

Interactive version: theaggregate.ai/benchmark?slug=findit-referring-expression-detection · How It Works · Data refreshed daily, snapshot 2026-09-29.