Vision & documents

RefCOCO (avg) leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on RefCOCO (avg) is 92.5% by Qwen3.6-27B. 6 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.6 Plus	92.6%	estimated ± 1.8 pp, low confidence
2	Qwen3.6-27B	92.5%	measured
3	Qwen3.5-122B-A10B	92.4%	estimated ± 1.8 pp, low confidence
4	Qwen3.5-27B	92.2%	estimated ± 1.8 pp, medium confidence
5	Qwen3.5-35B-A3B	92.1%	estimated ± 1.8 pp, medium confidence
6	Qwen3.6-35B-A3B	92.0%	measured
7	Command A+	91.4%	estimated ± 1.8 pp, medium confidence
8	Nemotron 3 Nano Omni 30B A3B	90.5%	measured
9	LFM2.5-VL-3B	87.9%	measured
10	ZAYA1-VL-8B	84.3%	measured
11	Interfaze Beta	82.1%	measured
12	LFM2.5-VL-450M	80.3%	estimated ± 1.8 pp, low confidence

All leaderboards

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General