AgentBench: Evaluating LLMs as Agents

TL;DR

AgentBench is a comprehensive benchmark for evaluating large language models as agents across diverse environments, revealing performance gaps and guiding improvements in reasoning, decision-making, and instruction following.

Contribution

The paper introduces AgentBench, a multi-dimensional benchmark with 8 environments for assessing LLMs as agents, and provides extensive evaluation results and insights into their strengths and weaknesses.

Findings

Top commercial LLMs excel in complex environments.

OSS LLMs underperform compared to commercial counterparts.

Poor long-term reasoning and instruction following are key challenges.

Abstract

The potential of Large Language Model (LLM) as agents has been widely acknowledged recently. Thus, there is an urgent need to quantitatively \textit{evaluate LLMs as agents} on challenging tasks in interactive environments. We present AgentBench, a multi-dimensional benchmark that consists of 8 distinct environments to assess LLM-as-Agent's reasoning and decision-making abilities. Our extensive test over \num API-based and open-sourced (OSS) LLMs shows that, while top commercial LLMs present a strong ability of acting as agents in complex environments, there is a significant disparity in performance between them and many OSS competitors that are no larger than 70B. We identify the typical reasons of failures in environments and LLMs, showing that poor long-term reasoning, decision-making, and instruction following abilities are the main obstacles for developing usable LLM agents. Improving instruction following and training on high quality multi-round alignment data could improve agent performance. And different from existing assumptions, training on code present ambivalent impacts on different agent tasks. Datasets, environments, and an integrated evaluation package for AgentBench are released at https://github.com/THUDM/AgentBench.