Personalized Reward Modeling research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

PRISM Personalized

P-GenRM

68.06Accuracy

Feb 26, 2026

Chatbot Arena Personalized

P-GenRM

75.92Accuracy

Feb 26, 2026

BESPOKE-Meta OOD

P-CHECK

75.48Binary Preference Accuracy

Feb 26, 2026

Reddit TLDR 150 examples Overall

MRM

69.7User-level Accuracy

Feb 26, 2026

Reddit TLDR 150 examples Unseen

MRM

69.8User-level Accuracy

Feb 26, 2026

Reddit TLDR 150 examples Seen

MRM

69.7User-level Accuracy

Feb 26, 2026

Reddit TLDR 100 examples Overall

MRM

69.6User-level Accuracy

Feb 26, 2026

Reddit TLDR 100 examples Unseen

MRM

69.6User-level Accuracy

Feb 26, 2026

Reddit TLDR 100 examples Seen

MRM

69.6User-level Accuracy

Feb 26, 2026

PRISM Overall

MRM

65.3User-level Accuracy

Feb 26, 2026

PRISM Unseen

MRM

0.652User-level Accuracy

Feb 26, 2026

PRISM Seen

MRM

65.3User-level Accuracy

Feb 26, 2026

DSP

PAFO

78.08Accuracy

Jun 9, 2026

Personal-LLM

PAFO

88.65Accuracy

Jun 9, 2026

Lamp-QA (OOD)

Qwen3-235B-A22B

60Arts Score

Feb 26, 2026

LaMP 4

Probabilistic User RM

17.9ROUGE@N=30

May 13, 2026