Unverified60% confidenceFactExact time
TerminalBench是一个包含约82个计算机使用和编程任务的Agent评估框架
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
Pi编程Agent深度解析：为何抛弃主流AI工具从零自建
bilibili63号炼金工坊6/3/2026
Related Claims
UnverifiedTerminal Bench是斯坦福大学设计的智能体评估框架，模拟真实Linux终端环境，要求智能体完成环境配置、依赖安装、代码调试、结果验证等完整工作流68% similarUnverifiedTerminal Bench要求模型在交互式命令行环境中调试、运行、修复代码，区别于HumanEval等静态代码生成测试60% similarUnverified新兴智能体评估框架包括AgentBench、WebArena、SWE-bench，分别针对工具调用、网页操作、代码修复场景58% similarUnverifiedTerminal-Bench 要求模型在真实的 shell 环境中进行多轮交互，包括发出命令、解析 stdout/stderr 输出、处理权限问题、管理文件系统状态58% similarUnverifiedThe architecture of terminal AI coding tools typically includes a session management layer, a model invocation layer, a tool execution layer, and a permission control layer57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/43520API
curl https://kongchang.com/api/v1/knowledge/claims/43520MCP
get_claim(id=43520)