Model Ranking Agreement on MT-Bench Arena-derived reference ranking 1.0 (test) benchmark leaderboard · SOTA2 Research

Model Ranking Agreement on MT-Bench Arena-derived reference ranking 1.0 (test)

121Coding ScoreSample25→20

Evaluation Results

Method	Links
Sample25→202026.06		121	128	121	122	132	80	95	168	121
Single→202026.06		83	74	54	39	55	57	54	168	73
Single→202026.06		83	111	86	49	98	87	113	132	95
ScoreWin402026.06		79	75	36	25	50	59	52	141	65
Sample25→202026.06		78	182	126	99	111	99	102	174	122
Random50→20×102026.06		76	104	46	30	54	59	48	145	70
NoTrunc boot2026.06		75	99	44	28	51	57	46	142	68
BlockTop22026.06		73	105	53	32	37	65	51	142	70
Trunc25→202026.06		70	49	61	24	54	44	47	102	56
Random50→20×102026.06		69	132	90	43	62	80	60	162	87
NoTrunc boot2026.06		66	118	88	41	57	79	59	159	83
BlockTop12026.06		65	114	59	29	44	43	83	139	72
ScoreWin402026.06		60	84	85	35	51	69	55	152	74
BlockTop12026.06		54	137	91	30	70	93	56	159	86
BlockTop22026.06		53	94	86	20	54	92	53	156	76
Trunc25→202026.06		52	55	90	20	56	53	72	159	70