Interactive Visual Grounding (Partial): leaderboard
Metric: Matcher accuracy (%; share of target items the model selects correctly given only the least informative attribute of the description, so the Matcher must ask for the rest; 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 |
| 2 | GPT-5.5 | 71.4 |
| 3 | GPT-5.4 (2026-03-05) | 65.5 |
| 4 | Gemma 4 31B (IT) | 58.4 |
| 5 | Gemini 3.1 Flash Lite (Preview) | 54.5 |
| 6 | Qwen 3 VL 32B Instruct | 44 |
| 7 | GPT-5.4 Mini | 37.1 |
| 8 | Qwen 3 VL 8B Instruct | 35.3 |
Interactive version: theaggregate.ai/benchmark?slug=interactive-visual-grounding-partial · How It Works · Data refreshed daily, snapshot 2026-09-29.