Interactive Visual Grounding (Static): leaderboard
Metric: Matcher accuracy (%; share of target items the model selects correctly given the full human-produced description, no interaction; 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) | 87.5 |
| 2 | Gemini 3.1 Flash Lite (Preview) | 67.8 |
| 3 | GPT-5.4 (2026-03-05) | 66.1 |
| 4 | GPT-5.5 | 65.8 |
| 5 | Gemma 4 31B (IT) | 56 |
| 6 | Qwen 3 VL 32B Instruct | 48.8 |
| 7 | GPT-5.4 Mini | 47.9 |
| 8 | Qwen 3 VL 8B Instruct | 33.6 |
Interactive version: theaggregate.ai/benchmark?slug=interactive-visual-grounding-static · How It Works · Data refreshed daily, snapshot 2026-09-29.