τ-Bench

Ψ-Bench is a benchmark designed to evaluate the ability of language models to influence users through persuasive dialogues by incorporating personalized user profiles derived from dialogue histories.

Papers using τ-Bench (8)

Goal Alignment in LLM-Based User Simulators for Conversational AI2025 · 14 citesReinforcement World Model Learning for LLM-based Agents2026LRanker: LLM Ranker for Massive Candidates2026Uncertainty-Aware Clarification in LLM Agents with Information Gain2026π-Bench: Evaluating Proactive Personal Assistant Agents in Long-Horizon Workflows2026Ψ-Bench: Evaluating Persona-Sensitive Influencing in Persuasive Dialogues2026Learning Agentic Policy from Action Guidance2026Planner-R1: Reward Shaping Enables Efficient Agentic RL with Smaller LLMs2025