List your company

ResearchFormerly WizWand

Step-level Correctness Discrimination on ProcessBench (GSM8K, MATH, Olympiad Bench, Omni Math)

0.242GSM8K Error RateRLHFlow-PRM-Deepseek-8B

Evaluation results

Method	Links
RLHFlow-PRM-Deepseek-8B2026.01		0.242	0.984	0.388	0.214	0.8	0.338	0.101	0.51	0.169	0.109	0.519	0.169	0.266
Math-Shepherd-PRM-7B2026.01		0.324	0.917	0.479	0.18	0.82	0.295	0.15	0.711	0.248	0.142	0.73	0.238	0.315
Qwen2.5-Math-7B-MC2026.01		0.332	0.927	0.487	0.215	0.778	0.325	0.126	0.601	0.206	0.129	0.546	0.199	0.304
RLHFlow-PRM-Mistral-8B2026.01		0.338	0.99	0.504	0.217	0.722	0.334	0.082	0.431	0.138	0.096	0.452	0.158	0.284
EurusPRM-Stage12026.01		0.469	0.42	0.443	0.333	0.382	0.356	0.239	0.198	0.217	0.219	0.245	0.231	0.312
EurusPRM-Stage22026.01		0.512	0.44	0.473	0.364	0.35	0.357	0.257	0.18	0.212	0.231	0.191	0.209	0.313
Qwen2.5-Math-7B-MCRD2026.01		0.556	0.945	0.676	0.419	0.785	0.521	0.225	0.612	0.288	0.208	0.557	0.285	0.443
Skywork-PRM-Qwen2.5-7B2026.01		0.618	0.829	0.708	0.438	0.622	0.536	0.179	0.319	0.229	0.14	0.419	0.21	0.421
Qwen2.5-72B-Instruct2026.01		0.628	0.969	0.762	0.463	0.931	0.618	0.387	0.926	0.546	0.366	0.909	0.522	0.612
Qwen2.5-Math-7B-NAIT2026.01		0.655	0.951	0.779	0.552	0.798	0.646	0.358	0.605	0.441	0.329	0.619	0.428	0.574
GPT-4o-08062026.01		0.7	0.912	0.792	0.544	0.766	0.636	0.458	0.584	0.514	0.452	0.656	0.535	0.619
Qwen2.5-Math-PRM-7B2026.01		0.72	0.964	0.824	0.68	0.904	0.776	0.557	0.855	0.675	0.552	0.83	0.663	0.735