OpenRef - Polysemy: leaderboard
The polysemy split of OpenRef (2026), an open-world referring expression benchmark: expressions built on polysemous words, to be grounded with bounding boxes in the image. The score is F1 over predicted and ground-truth boxes in percent; higher is better.
Metric: Grounding F1 (%) on expressions built on polysemous words, 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 | 71.4 | #927 |
| 2 | Qwen 3 VL 4B | 68.6 | #1163 |
| 3 | GLM-4.6V-Flash | 68.2 | #1049 |
| 4 | MiMo-VL-7B-RL | 65.5 | #1160 |
| 5 | Qwen 2.5 VL 7B | 64.5 | #1878 |
| 6 | Keye-VL-1.5-8B | 64.4 | #1729 |
| 7 | Qwen 3 VL 2B | 63.5 | #2156 |
| 8 | Qwen 2.5 VL 32B | 61.4 | #944 |
| 9 | LLaVA-OneVision-1.5-8B | 59.1 | #1876 |
| 10 | MiniCPM-V-4.5 | 54.9 | #1451 |
| 11 | InternVL3.5-8B | 52.1 | #1618 |
| 12 | Qwen 2 VL 7B | 46.8 | #2300 |
| 13 | OpenRef Mistral-3 8B (checkpoint unspecified) | 4.6 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=openref-polysemy · How It Works · Data refreshed daily, snapshot 2026-11-22.