Microsoft Research and UC Berkeley researchers release TokenEconomizer to reduce coding agent costsMachine translation
Microsoft Research 与 UC Berkeley 研究团队发布 TokenEconomizer，测试中降低编程智能体成本
Automatically verified and published · Generated and evidence-checked automatically; not reviewed by a human.
The open-source plugin uses records of past tasks in a repository to adjust prompts and tool settings for later tasks without changing model weights or agent code. Across 277 SWE-Bench tasks and five agent-model settings, mean cost fell by 18.4% to 30.9%; Pass@1 fell by 1.0 point in one setting.
This source permits summary display only.
Read at the original source