FrontierCode 2.0 released with broader tests of whether coding agents produce mergeable codeMachine translation
FrontierCode 2.0 发布，扩大代码代理可合并性评测
Automatically verified and published · Generated and evidence-checked automatically; not reviewed by a human.
FrontierCode 2.0 draws tasks from more repositories and applies stricter criteria for correctness, design and missed issues. Its publisher says the tasks come from 55% more repositories; in its results, the top-scoring model, Opus 5.5, reaches only 45%, leaving room for coding agents to improve on real engineering tasks.
This source permits summary display only.
Read at the original source