Stanford UniversityAI Measurement ScienceAIMS — AI Measurement Science

Reward Modeling

RewardBench

RewardBench per-(reward-model, item) correctness on chosen>rejected ranking.

2,733items
118subjects
ODC-BYlicense
reward_modelingdomain
textmodality
item-level responses released
Saturation status: Unknown
Original sourcePaper← All benchmarks

Response matrix

Fit to width. Hover for subject & item; click a cell for details.

RewardBench response matrix: AI models (rows) against items (columns)
lowhighUnobserved

Scale: {0, 0.5, 1} (0.5 = judge tie, <1% of rows)

Subjects

  1. 1infly/INF-ORM-Llama3.1-70B0.9615
  2. 2nicolinho/QRM-Gemma-2-27B0.9518
  3. 3Skywork/Skywork-Reward-Gemma-2-27B-v0.20.9497
  4. 4Skywork/Skywork-Reward-Gemma-2-27B0.9451
  5. 5Skywork/Skywork-Reward-Llama-3.1-8B-v0.20.9404
  6. 6nicolinho/QRM-Llama3.1-8B-v20.939
  7. 7LxzGordon/URM-LLaMa-3.1-8B0.9374
  8. 8Skywork/Skywork-Reward-Llama-3.1-8B0.9333
  9. 9Ray2333/GRM-Llama3-8B-rewardmodel-ft0.924
  10. 10Ray2333/GRM-llama3.2-3B-rewardmodel-ft0.9209
  11. 11internlm/internlm2-20b-reward0.9176
  12. 12LxzGordon/URM-LLaMa-3-8B0.9142
  13. 13NCSOFT/Llama-3-OffsetBias-RM-8B0.9028
  14. 14internlm/internlm2-7b-reward0.8938
  15. 15Ray2333/GRM-llama3-8B-sftreg0.8908
  16. 16Skywork/Skywork-Critic-Llama-3.1-8B0.8898
  17. 17openai/gpt-4o-2024-08-060.8861
  18. 18Ray2333/GRM-gemma2-2B-rewardmodel-ft0.8858
  19. 19Ray2333/GRM-llama3-8B-distill0.8831
  20. 20openai/gpt-4o-2024-05-130.867
  21. 21opencompass/CompassJudger-1-14B-Instruct0.8665
  22. 22sfairXC/FsfairX-LLaMA3-RM-v0.10.8637
  23. 23meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo0.8613
  24. 24mattshumer/Reflection-70B0.8606
  25. 25meta-llama/Meta-Llama-3.1-70B-Instruct0.8586
  26. 26CIR-AMS/BTRM_Qwen2_7b_06130.8566
  27. 27berkeley-nest/Starling-RM-34B0.8509
  28. 28Ray2333/GRM-Gemma-2B-rewardmodel-ft0.8496
  29. 29opencompass/CompassJudger-1-7B-Instruct0.8479
  30. 30openbmb/Eurus-RM-7b0.8449
  31. 31google/gemini-1.5-flash-0010.8382
  32. 32allenai/tulu-v2.5-13b-preference-mix-rm0.8322
  33. 33NCSOFT/Llama-3-OffsetBias-8B0.8275
  34. 34openai/gpt-4o-mini-2024-07-180.8271
  35. 35weqweasdas/RM-Mistral-7B0.8235
  36. 36google/gemma-2-27b-it0.8216
Full data on Hugging FaceBack to the gallery