Semantic Watermark Evaluation on WMVLM Semantic Watermarks 1.0 (test) benchmark leaderboard · SOTA2 Research

Semantic Watermark Evaluation on WMVLM Semantic Watermarks 1.0 (test)

85.2RingID Quality AccuracyWMVLM

Evaluation Results

Method	Links
WMVLM2026.01		85.2	85.2	95.6	95.6	91	91
GPT-52026.01		21.4	21.4	0.6	0.6	55.8	55.8
Gemma-3-12B2026.01		6.8	7.4	71.2	70.6	8.6	8.8
Gemini-3-Pro2026.01		5.4	6	1.2	0.8	71.6	71.6
Qwen3-VL-8B2026.01		4.6	5.4	0.8	0.2	90.8	90.8
Qwen3-VL-32B2026.01		4.6	5.4	2	10.8	84.4	83
Claude-Opus-4.52026.01		4.2	13.8	32.4	27	57.2	57.6
Gemma-3-27B2026.01		4.2	6.4	41.4	41	1.8	1.4
Qwen3-VL-4B2026.01		1.4	2.2	0.2	0.2	89	89
Gemma-3-4B2026.01		0.6	0.6	24.6	24.6	7.8	14
LLaVA-v1.6-7B2026.01		0	0.2	0.6	0.8	0.8	0.8
LLaVA-v1.6-13B2026.01		0	0	0	0.2	8.2	8.2