List your company

Helpfulness Preference Labeling Accuracy on Ultra-Real

74.8Benchmark ScoreFew-Shot
68.87270.41171.9573.489Feb 14, 2026

Evaluation results

MethodLinks
Few-ShotJudge=Deepseek-V32026.0274.857.317.5
Few-ShotJudge=Qwen3-14B2026.0274.156.517.6
RandomJudge=Qwen3-14B2026.0273.457.216.2
SeedJudge=Deepseek-V32026.0273.461.112.3
RandomJudge=Deepseek-V32026.0273.459.713.7
Biased Rubric SearchJudge=Qwen3-14B2026.0273.252.420.8
Few-ShotJudge=Gemma-3-27b-it2026.0272.96111.9
SeedJudge=Qwen3-14B2026.0272.861.910.9
TextGradJudge=Deepseek-V32026.0272.153.918.2
TextGradJudge=Qwen3-14B2026.0271.957.914
Biased Rubric SearchJudge=Deepseek-V32026.0271.954.117.8
TextGradJudge=Gemma-3-27b-it2026.0271.662.19.5
RandomJudge=Gemma-3-27b-it2026.0271.5647.5
SeedJudge=Gemma-3-27b-it2026.0270.263.56.7
Biased Rubric SearchJudge=Gemma-3-27b-it2026.0269.158.310.8