OpenRef - Logo: leaderboard

The logo split of OpenRef (2026), an open-world referring expression benchmark: proper-noun logo 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 logo 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 3 VL 8B57#917
2Qwen 3 VL 4B56.8#1151
3GLM-4.6V-Flash55.9#1040
4Qwen 3 VL 2B34.9#2134
5MiniCPM-V-4.533.5#1441
6Qwen 2.5 VL 32B31.5#941
7Qwen 2.5 VL 7B27.6#1862
8Qwen 2 VL 7B21.7#2275
9Keye-VL-1.5-8B12.4#1720
10MiMo-VL-7B-RL8.9#1155
11LLaVA-OneVision-1.5-8B7.6#1864
12InternVL3.5-8B2.9#1604
13OpenRef Mistral-3 8B (checkpoint unspecified)0.7

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

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