ResearchBenchmarksRubric Generation on OpenAI HealthBench OSS EVAL-300Rubric Generation on OpenAI HealthBench OSS EVAL-300Follow4.9BLEURubricRAG1.8842.6673.454.233Mar 21, 2026Evaluation resultsMethodMethodLinksBLEUROUGE-1ROUGE-2ROUGE-LLLM Judge ScoreRubricRAGThinking=nothinkThinking=nothink2026.034.933.111.526.90.567SFTMode=SFTMode=SFT2026.034.231.110.325.30.481RubricRAGThinking=thinkThinking=think2026.033.730.49.724.50.558GRPOMode=GRPOMode=GRPO2026.033.531.110.124.80.551Few-ShotMode=Few-ShotMode=Few-Shot2026.033.329.39.123.70.548Zero-ShotMode=Zero-ShotMode=Zero-Shot2026.03223.16.519.20.521