Unreal Labs推出了Unreal Agent，一个旨在降低生产环境成本的AI代理框架1。该框架通过异步工具调用管理机制，在生产工作负载上相比Codex实现40%的成本节省，相比Pi则节省20%1。
Unreal Agent的核心创新在于其事件日志机制处理工具调用，这一设计允许用户实现无延迟干预，同时支持多个工具任务的并行执行1。该框架已在多个编码基准测试中表现良好，并采用GPT-6 Astra xhigh进行测试1。为了方便开发者使用，Unreal Labs提供了Go库、Runner可执行文件以及Harbor兼容的基准测试运行器等技术支持1。
Unreal Labs has unveiled Unreal Agent, an artificial intelligence framework designed to optimize operational efficiency through improved tool management1. The system leverages asynchronous tool invocation to deliver substantial cost reductions compared to existing alternatives, achieving 40% savings relative to Codex and 20% savings versus Pi on production workloads1.
The framework's core innovation addresses a critical challenge in AI agent design: managing the overhead associated with tool interactions1. By implementing an event-logging mechanism for tool calls, Unreal Agent enables users to intervene in agent operations without latency and execute multiple tool tasks in parallel1. The system has been validated across multiple coding benchmarks using the GPT-6 Astra xhigh model1.
To support adoption, Unreal Labs provides developers with a Go library, a Runner executable, and a Harbor-compatible benchmark runner for performance testing1. Interested parties can reach the team at [email protected]1.
评论
还没有评论，欢迎留下第一条。