PieArena: Frontier Language Agents Achieve MBA-Level Negotiation Performance and Reveal Novel Behavioral Differences
Chris Zhu, Sasha Cui, Will Sanok Dufallo, Runzhi Jin, Zhen Xu, Linjun Zhang, Daylian Cain
TL;DR
PieArena introduces a large-scale negotiation benchmark that evaluates language models (LMs) and humans across realistic MBA-style scenarios with deterministic outcomes and cross-play, revealing that frontier LMs can match or exceed trained students in MBA-level negotiations. It combines a diverse dataset (326 models, 167 humans, >25k transcripts) with a Gaussian--Generalized Bradley--Terry--Luce ranking model and an agentic harness that implements shared intentionality to measure both performance and nuanced behaviors such as deception, computation accuracy, and reputation. The results show that agentic scaffolding provides large gains for mid- and lower-tier models, while frontier models exhibit diminishing returns; in inter-species tasks, GPT-5 often outperforms humans on single-issue bargains and remains highly competitive in multi-issue settings. Beyond outcomes, PieArena offers multi-dimensional capability profiles and regression analyses linking behavioral traits to value capture, highlighting robustness, alignment, and deployment considerations for real-world use. Overall, the work argues that frontier language agents are already capable of deployment in high-stakes economic settings, while emphasizing that trustworthiness and robustness remain critical challenges to address.
Abstract
We present an in-depth evaluation of LLMs' ability to negotiate, a central business task that requires strategic reasoning, theory of mind, and economic value creation. To do so, we introduce PieArena, a large-scale negotiation benchmark grounded in multi-agent interactions over realistic scenarios drawn from an MBA negotiation course at an elite business school. We find systematic evidence of AGI-level performance in which a representative frontier agent (GPT-5) matches or outperforms trained business-school students, despite a semester of general negotiation instruction and targeted coaching immediately prior to the task. We further study the effects of joint-intentionality agentic scaffolding and find asymmetric gains, with large improvements for mid- and lower-tier LMs and diminishing returns for frontier LMs. Beyond deal outcomes, PieArena provides a multi-dimensional negotiation behavioral profile, revealing novel cross-model heterogeneity, masked by deal-outcome-only benchmarks, in deception, computation accuracy, instruction compliance, and perceived reputation. Overall, our results suggest that frontier language agents are already intellectually and psychologically capable of deployment in high-stakes economic settings, but deficiencies in robustness and trustworthiness remain open challenges.