ResearchBenchmarksPersonalized Interaction on ultrachat SyntheticFollow7.26Personalization ScorePPOpt5.4925.9516.416.869Feb 12, 2026Evaluation ResultsMethodMethodLinksPersonalization ScoreTask Completion ScorePPOptBackbone=GPT-oss-20bBackbone=GPT-oss-20b2026.027.267.96PPOptBackbone=Llama-3-8b-in...Backbone=Llama-3-8b-instruct2026.0277.7PPOptBackbone=Qwen3-8bBackbone=Qwen3-8b2026.0277.82VanillaBackbone=Llama-3-8b-in...Backbone=Llama-3-8b-instruct2026.025.567.88VanillaBackbone=Qwen3-8bBackbone=Qwen3-8b2026.025.567.88VanillaBackbone=GPT-oss-20bBackbone=GPT-oss-20b2026.025.567.88