Behavioral Similarity Analysis on τ-Bench and τ2-Bench (test) benchmark leaderboard · SOTA2 Research

Behavioral Similarity Analysis on τ-Bench and τ2-Bench (test)

82.6GED ScoreSonnet 4.5

Evaluation Results

Method	Links
Sonnet 4.52026.04		82.6	3.14	37.1	95.1	79.1	79.9	94	84.3	4.07	3.89	3.66	3.87
Opus 4.12026.04		79.7	3.25	35.5	94.6	81	74.4	93.7	83	4.02	3.8	3.72	3.85
Kimi-K22026.04		78.1	2.81	34.3	93.8	82.6	70.8	94.7	82.7	3.86	3.57	3.51	3.65
Gemini 3 Pro2026.04		77.5	2.2	29.4	92.4	71.9	71.2	92.3	78.5	2.68	2.31	2.81	2.6
GLM-4.62026.04		75.9	2.71	30.6	92.5	80.4	71.9	88.7	80.3	3.45	3.26	3.54	3.42
GPT-4.12026.04		75.2	2.45	30.6	93.6	75.9	74.6	88	79.5	3.07	2.92	3.45	3.15
Kimi-K22026.04		74.8	2.79	31.8	92.5	70.2	73.3	90.4	77.9	3.67	3.33	3.38	3.46
Doubao 1.62026.04		72.9	2.24	27.6	93.1	76.9	67.3	88.3	77.5	2.71	2.56	3.38	2.88
V3.12026.04		72.6	2.56	30.7	93.2	78.1	63.5	91.2	77.6	3.24	2.91	3.43	3.19
Doubao 1.62026.04		71.8	2.07	27.4	92.8	76.7	68	88.8	77.8	2.59	2.47	3.33	2.8
R12026.04		70.8	2.44	25.9	92.5	78.3	72.5	85	78.6	2.99	2.87	3.29	3.05
Doubao 1.62026.04		70.3	2.32	28.7	93	68.5	69.5	88.1	75.4	2.69	2.57	3.36	2.87
Gemini 2.5 Pro2026.04		70.1	2.24	28.1	92.2	71.7	64.6	80.5	72.3	2.65	2.42	3.18	2.75
V3.12026.04		69.2	2.29	31.2	93.1	72.4	65.3	87.5	75.1	3.55	3.27	3.38	3.4
GPT-52026.04		68.3	2.26	23	88.9	71.3	69.4	87.7	76.1	2.5	2.43	3.17	2.7
Qwen3-30B2026.04		65.8	2.51	25	89.5	77.9	62.4	89.9	76.7	2.65	2.42	2.76	2.42
Qwen3-235B2026.04		65.6	2.05	24.5	89.7	68.1	67.2	92.4	75.9	2.62	2.43	2.77	2.4
V3-03242026.04		59.1	1.87	27	91.7	77.5	62.7	74.5	71.5	2.84	2.66	3.13	2.87