List your company
ResearchFormerly WizWand

Reward Modeling on PPE-Preference

79.8AccuracySkywork-Reward-V2-Llama-3.1-8B-40M
51.258.62566.0573.475Jul 2, 2025Aug 26, 2025Oct 20, 2025Dec 14, 2025Feb 7, 2026Apr 3, 2026May 29, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Skywork-Reward-V2-Llama-3.1-8B-40MParadigm=Our Reward Models2025.0779.8——
Skywork-Reward-V2-Llama-3.1-8BParadigm=Our Reward Models2025.0777.3——
Skywork-Reward-V2-Qwen3-8BParadigm=Our Reward Models2025.0770.6——
Gemini-2.5-ProCategory=LLM-as-Judge, Tier=Pro2026.0269.88.364
Skywork-Reward-V2-Qwen3-4BParadigm=Our Reward Models2025.0769.5——
Claude-Sonnet-4.5Category=LLM-as-Judge2026.0269.214.559.1
Skywork-Reward-V2-Llama-3.2-3BParadigm=Our Reward Models2025.0769.1——
Claude-Sonnet-4.5-thinkingCategory=LLM-as-Judge, Mode=thinking2026.0268.78.762.7
SAVEPolicy Model=Qwen3-4B-Instruct-25072026.0568.6——
Continual Offline Training RM2026.0568.5——
Skywork-Reward-V2-Llama-3.2-3B2026.0568.4——
HL-BTPolicy Model=Qwen2.5-3B-Instruct2026.0568.4——
HL-BTPolicy Model=Qwen3-4B-Instruct-25072026.0568.3——
SAVE (w/o Policy Model Optimization)Policy Model=Qwen2.5-3B-Instruct2026.0567.9——
SAVE (w/o Curriculum Mechanism)Policy Model=Qwen3-4B-Instruct-25072026.0567.8——
SAVE (w/o Policy Model Optimization)Policy Model=Qwen3-4B-Instruct-25072026.0567.8——
GPT-4oParadigm=LLM-as-a-Judge & Generative Reward Models2025.0767.7——
Skywork-Reward-V2-Qwen3-1.7BParadigm=Our Reward Models2025.0767.6——
SAVEPolicy Model=Qwen2.5-3B-Instruct2026.0567.5——
SAVE (w/o Curriculum Mechanism)Policy Model=Qwen2.5-3B-Instruct2026.0567.5——
Claude-3.5-SonnetParadigm=LLM-as-a-Judge & Generative Reward Models2025.0767.3——
DeepSeek-GRM-27BParadigm=LLM-as-a-Judge & Generative Reward Models, Integration=w/ MetaRM2025.0767.2——
J1-Llama-70BParadigm=LLM-as-a-Judge & Generative Reward Models, Sampling Strategy=Maj@322025.0767——
DeepSeek-V3.2-thinkingCategory=LLM-as-Judge, Mode=thinking2026.0266.826.249.2
GPT-5-chatCategory=LLM-as-Judge, Mode=chat2026.0266.619.653.5
Skywork-Reward-V2-Llama-3.2-1BParadigm=Our Reward Models2025.0766.6——
DeepSeek-V3.2-chatCategory=LLM-as-Judge, Mode=chat2026.0266.42450.5
J1-Llama-70BParadigm=LLM-as-a-Judge & Generative Reward Models2025.0766.3——
GenRM-RLVR-14BCategory=Our Methods (Baseline), Parameters=14B, Training=RLVR2026.0265.745.735.6
GenRM-R-Align-14BCategory=Our Methods, Parameters=14B, Training=R-Align2026.0265.726.948.1
RM-R1-Qwen-32BCategory=Specialized Generative Reward Models, Parameters=32B2026.0265.63045.9
DeepSeek-GRM-27BParadigm=LLM-as-a-Judge & Generative Reward Models2025.0765.3——
Skywork-Reward-V2-Qwen3-0.6BParadigm=Our Reward Models2025.0765.3——
Qwen3-14BCategory=Our Methods (Baseline), Parameters=14B2026.0265.236.941.1
RRM-32BCategory=Specialized Generative Reward Models, Parameters=32B2026.0265.162.424.5
RM-R1-DS-32BCategory=Specialized Generative Reward Models, Parameters=32B2026.0264.946.534.7
GPT-5-thinkingCategory=LLM-as-Judge, Mode=thinking2026.0264.76.958.8
GPT-OSS-120BCategory=LLM-as-Judge, Parameters=120B2026.0264.540.738.2
Llama-3.1-Nemotron-70BParadigm=Bradley-Terry2025.0764.2——
INF-ORM-Llama3.1-70BParadigm=Bradley-Terry2025.0764.2——
GenRM-RLVR-8BCategory=Our Methods (Baseline), Parameters=8B, Training=RLVR2026.0263.747.433.5
GenRM-R-Align-8BCategory=Our Methods, Parameters=8B, Training=R-Align2026.0263.634.941.4
Skywork-Reward-Gemma-2-27B-v0.2Paradigm=Bradley-Terry2025.0763.6——
GRM-llama3-8B-distillParadigm=Bradley-Terry2025.0763.2——
FsfairX-LLaMA3-RM-v0.1Paradigm=Bradley-Terry2025.0763.1——
Starling-RM-34BParadigm=Bradley-Terry2025.0762.8——
Gemini-3-ProCategory=LLM-as-Judge, Tier=Pro2026.0262.71.661.6
LDL-Reward-Gemma-2-27B-v0.1Paradigm=Bradley-Terry2025.0762.4——
Mean RewardPolicy Model=Qwen2.5-3B-Instruct2026.0562.4——
Qwen3-8BCategory=Our Methods (Baseline), Parameters=8B2026.0262.245.234.1
Skywork-Reward-Llama-3.1-8B-v0.2Paradigm=Bradley-Terry2025.0762.2——
Internlm2-7b-rewardParadigm=Bradley-Terry2025.0762.1——
GRM-Llama3-8B-rewardmodel-ftParadigm=Bradley-Terry2025.0762.1——
Skywork-Reward-Llama-3.1-8BParadigm=Bradley-Terry2025.0762.1——
RM-Mistral-7BParadigm=Bradley-Terry2025.0761.8——
BTRM Qwen2 7b 0613Paradigm=Bradley-Terry2025.0761.8——
Qwen3-4B-Thinking-2507Category=LLM-as-Judge, Parameters=4B, Mode=Thinking2026.0261.533.740.8
Internlm2-20b-rewardParadigm=Bradley-Terry2025.0761——
ArmoRM-Llama3-8B-v0.1Paradigm=Bradley-Terry2025.0760.6——
QRM-Llama3.1-8BParadigm=Bradley-Terry2025.0760.6——
J1-Llama-8BParadigm=LLM-as-a-Judge & Generative Reward Models, Sampling Strategy=Maj@322025.0760.6——
Qwen3-4B-Instruct-2507Category=LLM-as-Judge, Parameters=4B, Mode=Instruct2026.0260.444.833.3
Skywork-Reward-Gemma-2-27BParadigm=Bradley-Terry2025.0760.3——
J1-Llama-8BParadigm=LLM-as-a-Judge & Generative Reward Models2025.0760.3——
URM-LLaMa-3.1-8BParadigm=Bradley-Terry2025.0760.2——
Mean RewardPolicy Model=Qwen3-4B-Instruct-25072026.0560——
GRM-gemma2-2B-rewardmodel-ftParadigm=Bradley-Terry2025.0759.7——
Eurus-RM-7bParadigm=Bradley-Terry2025.0759.6——
Llama-3-OffsetBias-RM-8BParadigm=Bradley-Terry2025.0759.2——
internlm2-1.8b-rewardParadigm=Bradley-Terry2025.0757.3——
QRM-Llama3.1-8B-v2Paradigm=Bradley-Terry2025.0757.2——
QRM-Gemma-2-27BParadigm=Bradley-Terry2025.0752.3——