SG-DT parsing on NormBench (Random split) benchmark leaderboard · SOTA2 Research

SG-DT parsing on NormBench (Random split)

100Done RateHeuristic Attach

Evaluation Results

Method	Links
Heuristic Attach2026.06		100	17.39	8.74	0.8409	100	0	62.78	25.6	83.83	1.14
Neural Biaffine Parser2026.06		100	19.26	9.37	0.8287	100	0	67.2	28.13	84.01	2.27
Constrained Decoder2026.06		100	19.26	9.29	0.8294	100	0	67.17	28.02	84.01	2.27
GPT-5.2 (2025-12-11)2026.06		84.11	42.04	21.5	0.6866	79.29	20.71	62.91	47.06	77.57	56.87
Gemini-3-Pro-preview2026.06		82.81	45.19	24.62	0.6572	77.28	22.72	64.34	50.28	76.04	54.52
Qwen3-235B-A22B-thinking2026.06		81.78	40.9	20.16	0.7031	73.39	26.52	62.39	47.29	74.31	48.79
DeepSeek-V3.22026.06		78.62	42.34	23.01	0.6788	73.5	26.41	59.73	47.51	73.15	61.5
Claude-Opus-4.52026.06		76.67	42.85	22.83	0.6806	70.7	29.11	59.15	47.8	71.52	64
LawGPT2026.06		76.3	15.77	4.81	0.8965	60.56	37.21	41.22	20.74	59.36	7.36
GLM-4.72026.06		74.72	38.56	20.72	0.7087	68.71	31.2	56.47	42.51	67.58	47.35
ChatLaw2026.06		72.49	18.04	4.65	0.8896	69.26	30.74	43.34	22.55	57.9	0.65
LegalOne-8B2026.06		65.15	15.14	2.66	0.9222	54.55	44.8	35.04	15.82	54.53	29.39