Large Language Model Watermarking on BOOKSUM (test) benchmark leaderboard · SOTA2 Research

Large Language Model Watermarking on BOOKSUM (test)

4.38Average RankSynthID

Evaluation Results

Method	Links
SynthID2026.05		4.38	99.8	31.4	15.6	21.2	47.8	33.4	18.8	-0.67
PMark2026.05		4.61	98	51.5	57.5	77	57.2	37.6	5.2	-1.15
k-SemStamp2026.05		4.94	98.6	61.7	43.1	65.9	51.4	47.2	1.4	-0.82
SemStamp2026.05		5.04	96	49.8	39.8	58.6	53.2	39.2	7.6	-0.72
None2026.05		5.06	—	—	—	—	—	—	—	0
SynthID2026.05		5.11	99.8	23.8	10.4	12.6	54.2	38.6	7.2	-0.65
SAMARK2026.05		5.14	98	90.5	88.1	87.1	47.9	45.2	6.9	-0.62
SemStamp2026.05		5.24	97	46.9	34.5	64.3	39.2	41.6	19.2	0.18
UPV2026.05		5.42	99.4	71	48.8	61.6	39.2	38.2	22.6	0.36
EXP2026.05		5.56	99.8	38.2	17.2	29.6	52.4	46.2	1.4	-0.2
MorphMark2026.05		5.57	100	74	46.2	64	36.4	40.4	23.2	0.51
k-SemStamp2026.05		5.58	97.6	51.7	37.1	63.7	35	45	20	0.52
SAMARK2026.05		5.75	98.8	92.6	90.2	88.2	34.2	46	19.8	0.69
None2026.05		5.76	—	—	—	—	—	—	—	0
EXP2026.05		5.81	99.8	42.2	16.2	37.8	30.6	45.8	23.6	0.75
KGW2026.05		5.97	100	76.8	45.6	66.8	32	42.4	25.6	0.91
MorphMark2026.05		6.08	99.4	56	34.2	45.8	47.6	51.4	1	0.32
UPV2026.05		6.12	99	61.4	23.6	38	47.2	51.4	1.4	0.35
PMark2026.05		6.22	96.6	24.2	28.3	46.3	26	49.8	24.2	1.16
KGW2026.05		6.65	100	64	34.4	54.4	38	50.2	11.8	0.89
SAMARK2026.05		—	98.7	88.7	89	90.3	33	59	8	—
SAMARK2026.05		—	99.4	86.6	84.8	91.2	20.4	56.4	23.2	—