List your company

ResearchFormerly WizWand

Preference Classification

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

Anthropic HH Harmless (test)

71.7Accuracy

22

Feb 26, 2026

35.2Error Rate

7

Apr 7, 2026

WebGPT comparisons (test)

60.8Accuracy

7

Feb 26, 2026

Anthropic HH Helpful (test)

57.6Accuracy

7

Feb 26, 2026

Math Reasoning (test)

85.4Classification Accuracy

4

Mar 11, 2026

Instruction Following (IF) (test)

61.4Accuracy

4

Mar 11, 2026

Helpfulness & Harmlessness (HH) (test)

70.4Classification Accuracy

4

Mar 11, 2026