OpenRef - Celebrity: leaderboard

The celebrity split of OpenRef (2026), an open-world referring expression benchmark: proper-noun celebrity expressions, 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 proper-noun celebrity expressions, 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

#ModelScoreOverall rank
1Qwen 2.5 VL 32B91.4#941
2Qwen 3 VL 8B90.8#917
3Qwen 3 VL 2B89.1#2134
4Qwen 3 VL 4B89#1151
5Qwen 2.5 VL 7B85.6#1862
6MiniCPM-V-4.584.5#1441
7GLM-4.6V-Flash83.5#1040
8InternVL3.5-8B77#1604
9Keye-VL-1.5-8B77#1720
10LLaVA-OneVision-1.5-8B75.7#1864
11MiMo-VL-7B-RL70.5#1155
12Qwen 2 VL 7B59.3#2275
13OpenRef Mistral-3 8B (checkpoint unspecified)24.8

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=openref-celebrity · How It Works · Data refreshed daily, snapshot 2026-11-21.