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
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 89.9 |
| 2 | GPT-5.5 | 69.5 |
| 3 | GPT-5.4 (2026-03-05) | 69.4 |
| 4 | Gemini 3.1 Flash Lite (Preview) | 65.3 |
| 5 | Gemma 4 31B (IT) | 60 |
| 6 | Qwen 3 VL 32B Instruct | 50.5 |
| 7 | GPT-5.4 Mini | 49 |
| 8 | Qwen 3 VL 8B Instruct | 35 |
Interactive version: theaggregate.ai/benchmark?slug=interactive-visual-grounding-full · How It Works · Data refreshed daily, snapshot 2026-09-29.