ResearchBenchmarksPersonalization on Real-world (test)Personalization on Real-world (test)Follow7.35ScorePPOpt5.41565.91786.426.9222Feb 12, 2026Evaluation resultsMethodMethodLinksScoreDelta ScorePPOptLLM Backbone=GPT-oss-20bLLM Backbone=GPT-oss-20b2026.027.351.86PPOptLLM Backbone=Llama-3-8...LLM Backbone=Llama-3-8b-instruct2026.027.21.71PPOptLLM Backbone=Qwen3-8bLLM Backbone=Qwen3-8b2026.027.11.61VanillaLLM Backbone=Llama-3-8...LLM Backbone=Llama-3-8b-instruct2026.025.49—VanillaLLM Backbone=Qwen3-8bLLM Backbone=Qwen3-8b2026.025.49—VanillaLLM Backbone=GPT-oss-20bLLM Backbone=GPT-oss-20b2026.025.49—