TRIP-Bench：现实场景中长期交互智能体基准测试框架

摘要

随着基于大语言模型的智能体被部署到日益复杂的现实场景中，现有基准测试未能充分体现关键挑战，例如执行全局约束、协调多工具推理，以及在长周期多轮交互中适应用户行为的动态变化。为弥补这一缺陷，我们推出了TRIP-Bench——一个基于真实旅行规划场景的长周期基准测试。该基准利用真实世界数据，提供18个精选工具和40余项旅行需求，并支持自动化评估。其包含不同难度的测试集：困难级测试集重点关注冗长模糊的交互、风格转换、可行性变更及迭代式方案修订。对话最长可达15轮用户交互，可能涉及150余次工具调用，上下文长度可超过20万词元。实验表明，即使先进模型在简单测试集上的最高成功率也仅达50%，而在困难子集上性能更降至10%以下。我们进一步提出GTPO方法——一种采用特定奖励归一化与奖励差分机制的在线多轮强化学习方法。将GTPO应用于Qwen2.5-32B-Instruct模型后，其在约束满足度与交互鲁棒性方面显著提升，在我们的评估中表现优于Gemini-3-Pro。我们期待TRIP-Bench能推动实用型长周期交互智能体的发展，而GTPO能为鲁棒的长周期训练提供有效的在线强化学习方案。

English

As LLM-based agents are deployed in increasingly complex real-world settings, existing benchmarks underrepresent key challenges such as enforcing global constraints, coordinating multi-tool reasoning, and adapting to evolving user behavior over long, multi-turn interactions. To bridge this gap, we introduce TRIP-Bench, a long-horizon benchmark grounded in realistic travel-planning scenarios. TRIP-Bench leverages real-world data, offers 18 curated tools and 40+ travel requirements, and supports automated evaluation. It includes splits of varying difficulty; the hard split emphasizes long and ambiguous interactions, style shifts, feasibility changes, and iterative version revision. Dialogues span up to 15 user turns, can involve 150+ tool calls, and may exceed 200k tokens of context. Experiments show that even advanced models achieve at most 50\% success on the easy split, with performance dropping below 10\% on hard subsets. We further propose GTPO, an online multi-turn reinforcement learning method with specialized reward normalization and reward differencing. Applied to Qwen2.5-32B-Instruct, GTPO improves constraint satisfaction and interaction robustness, outperforming Gemini-3-Pro in our evaluation. We expect TRIP-Bench to advance practical long-horizon interactive agents, and GTPO to provide an effective online RL recipe for robust long-horizon training.

TRIP-Bench：现实场景中长期交互智能体基准测试框架

TRIP-Bench: A Benchmark for Long-Horizon Interactive Agents in Real-World Scenarios

摘要

Support