ENTITY PROFILE / 实体档案
SWE-bench
SWE-bench is a benchmark for evaluating the performance of coding agents on software engineering tasks. The Agent Lightning v1.0 training pipeline raised Pass@1 on the SWE-bench Verified subset from 41.8 % to 56.4 %, a 14.6‑percentage‑point absolute gain, using only about 6 000 training samples. SWE-bench Pro provided data for Taste-Bench Flint leads coding benchmark SWE-Bench Pro v2· Graphiti 生成的实体摘要
RELATIONS
2关系CURRENT
2当前有效EVENTS
2相关事件REPORTS
2来源报道暂无可公开的邻近关系
按需展开实体的全部已公开关系，保留探索过的节点与连线。
关系变化
按生效时间- FlintLEADSSWE-bench
Flint leads coding benchmark SWE-Bench Pro v2
- SWE-benchPROVIDED DATA FORTaste-Bench
SWE-bench Pro provided data for Taste-Bench
关系是 Graphiti 从报道中提取的事实陈述，不是原文引述；失效时间表示后来的报道显示这条关系已不成立。