Step-wise Confidence Attribution on Math benchmark leaderboard · SOTA2 Research

Step-wise Confidence Attribution on Math

0.6946AUROCGIBS

Evaluation Results

Method	Links
GIBS2026.05		0.6946	0.8078	83.22	0.405
Cos-Mean2026.05		0.6116	0.8503	77.08	0.1013
NLI-Max2026.05		0.6043	0.8052	79.25	0.309
GIBS2026.05		0.5855	0.489	45.13	0.2737
GIBS2026.05		0.5806	0.6831	63.59	0.3786
NLI-Mean2026.05		0.5739	0.802	77.12	0.5935
NLI-Mean2026.05		0.5407	0.6694	62.67	0.495
Cos-Mean2026.05		0.5363	0.6583	61.77	0.2463
NLI-Max2026.05		0.531	0.6676	63.4	0.234
P(true)2026.05		0.5277	0.7435	75.54	0.7416
Entropy2026.05		0.519	0.6259	62.25	0.1814
Entropy2026.05		0.519	0.7351	74.79	0.2023
NLI-Max2026.05		0.5173	0.5061	47.43	0.1852
NLI-Mean2026.05		0.5148	0.4881	46.7	0.3356
SL(norm)2026.05		0.5138	0.4627	46.26	0.325
Entropy2026.05		0.512	0.4617	45.5	0.2876
Cos-Mean2026.05		0.5118	0.4848	46.07	0.3946
P(true)2026.05		0.5055	0.6298	62.26	0.6215
P(true)2026.05		0.4584	0.4166	45.51	0.4564
LECO2026.05		0.4378	0.4252	43.82	0.2229
Cos-Max2026.05		0.4313	0.4021	45.13	0.4408
SL(norm)2026.05		0.4093	0.5433	60.89	0.2319
LECO2026.05		0.4089	0.7279	70.44	0.3957
Cos-Max2026.05		0.3845	0.6617	76.56	0.1568
SL(norm)2026.05		0.3841	0.6597	73.4	0.222
LECO2026.05		0.3838	0.5836	59.01	0.1813
Cos-Max2026.05		0.3376	0.5015	59.37	0.3093