OpenRef: leaderboard
OpenRef (2026) tests open-world referring expression comprehension: expressions name landmarks, celebrities or logos, use polysemous or ordinal words, and may refer to one or several objects. The score is grounding F1 over predicted and ground-truth boxes in percent, averaged over its categories; higher is better.
Metric: Grounding F1 (%) over predicted and ground-truth boxes, average over the positive-expression categories, zero-shot box grounding, each model prompted in its own grounding format; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B | 63.7 | #917 |
| 2 | Qwen 3 VL 4B | 60.9 | #1151 |
| 3 | Qwen 2.5 VL 32B | 59.4 | #941 |
| 4 | GLM-4.6V-Flash | 58.9 | #1040 |
| 5 | Qwen 3 VL 2B | 54.2 | #2134 |
| 6 | Qwen 2.5 VL 7B | 52.6 | #1862 |
| 7 | MiniCPM-V-4.5 | 48.8 | #1441 |
| 8 | InternVL3.5-8B | 40.8 | #1604 |
| 9 | Keye-VL-1.5-8B | 40.7 | #1720 |
| 10 | MiMo-VL-7B-RL | 37.6 | #1155 |
| 11 | LLaVA-OneVision-1.5-8B | 35.3 | #1864 |
| 12 | Qwen 2 VL 7B | 34.1 | #2275 |
| 13 | OpenRef Mistral-3 8B (checkpoint unspecified) | 5.9 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=openref · How It Works · Data refreshed daily, snapshot 2026-11-21.