RefCOCO-avg: leaderboard
Average object-grounding score across RefCOCO, RefCOCO+, and RefCOCOg, measuring whether vision-language models localize referred objects.
Metric: Score (%). Source: github.com. Source: llm-stats.com. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 Plus | 93.5 | #120 |
| 2 | Qwen 3.6 27B | 92.5 | #170 |
| 3 | Qwen 3 VL 235B A22B (Thinking) | 92.4 | #238 (Qwen 3 VL 235B A22B) |
| 4 | Qwen 3.6 35B A3B | 92 | #226 |
| 5 | Qwen 3.5 122B A10B | 91.3 | #173 |
| 6 | Qwen 3.5 27B | 90.9 | #187 |
| 7 | Qwen 3.5 35B A3B | 89.2 | #256 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=refcoco-avg · How It Works · Data refreshed daily, snapshot 2026-10-20.