Interactive Visual Grounding (Full): leaderboard

Metric: Matcher accuracy (%; share of target items the model selects correctly given the full description with optional clarification questions; micro-averaged over 536 target items of four human-grounded image sets (dogs, two basket sets, tangrams), with GPT-5.4 as the simulated Director answering from the human description; a 10-turn budget per target, utterances capped at 50 words). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)89.9
2GPT-5.569.5
3GPT-5.4 (2026-03-05)69.4
4Gemini 3.1 Flash Lite (Preview)65.3
5Gemma 4 31B (IT)60
6Qwen 3 VL 32B Instruct50.5
7GPT-5.4 Mini49
8Qwen 3 VL 8B Instruct35

Interactive version: theaggregate.ai/benchmark?slug=interactive-visual-grounding-full · How It Works · Data refreshed daily, snapshot 2026-09-29.