List your company
ResearchFormerly WizWand

Preference Validation on RewardBench 2

74.5AccuracyMRRG
19.17233.53647.962.264Jul 2, 2026
Updated 4mo ago

Evaluation results

MethodLinks
MRRGBackbone=GPT-OSS-120B2026.0774.5
MRRGBackbone=GPT-OSS-20B2026.0773.7
MRRGBackbone=Qwen2.5-32B-Instruct2026.0765.3
SVRG w/ SRBackbone=GPT-OSS-120B2026.0764.4
SVRG w/ SRBackbone=GPT-OSS-20B2026.0762.4
Chasing the TailBackbone=GPT-OSS-120B2026.0761.3
Chasing the TailBackbone=GPT-OSS-20B2026.0760.3
Chasing the TailBackbone=Qwen2.5-32B-Instruct2026.0759.4
MRRGBackbone=Qwen2.5-7B-Instruct2026.0759.2
SVRG w/ SRBackbone=Qwen2.5-32B-Instruct2026.0754.9
SVRG w/o SRBackbone=GPT-OSS-120B2026.0752.3
MRRGBackbone=Qwen2.5-3B-Instruct2026.0751.2
SVRG w/o SRBackbone=GPT-OSS-20B2026.0751.2
SVRG w/ SRBackbone=Qwen2.5-7B-Instruct2026.0750.7
SVRG w/o SRBackbone=Qwen2.5-32B-Instruct2026.0742.3
SVRG w/ SRBackbone=Qwen2.5-3B-Instruct2026.0740.3
SVRG w/o SRBackbone=Qwen2.5-7B-Instruct2026.0737.6
Chasing the TailBackbone=Qwen2.5-7B-Instruct2026.0736.5
SVRG w/o SRBackbone=Qwen2.5-3B-Instruct2026.0729.4
Chasing the TailBackbone=Qwen2.5-3B-Instruct2026.0721.3