TVWorld-G: leaderboard
Focus-aware grounding on TV screens (TVWorld-G, Ma et al., 2026): given a Google TV screenshot, the model outputs the bounding box of the currently highlighted element. Acc@0.5 is the percent of 187 screenshots where the predicted box overlaps the annotated one with an IoU of at least 0.5; higher is better.
Metric: Acc@0.5 (%; predicted focused-element box with IoU of at least 0.5, 187 Google TV screenshots). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 78.1 | #982 |
| 2 | Qwen 3 VL 8B (Thinking) | 69.5 | #887 |
| 3 | Qwen 2.5 VL 7B Instruct | 66.3 | #1842 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=tvworld-g · How It Works · Data refreshed daily, snapshot 2026-11-12.