Auto-Rubric as Reward: From Implicit Preferences to Explicit Multimodal Generative Criteria

Benchmark Model Rank Results
image-editing-on-gedit-bench-enARR-RPO (Gemini 3.1 Pro-ARR)#7Overall: 7.85
text-to-image-generation-on-dpg-benchFLUX.1-dev + ARR-RPO (Gemini 3.1 Pro judge)#52DPG-Bench Score: 85.76
text-to-image-generation-on-genevalFLUX.1-dev + ARR-RPO (Gemini 3.1 Pro judge)#103Overall: 0.8