Published at 11/7/2026, 7:13:03 PM · 14 views
📰 Klavis Ships Multi-step SWE: 69 Tasks / 574 Dependent Steps; Vendor Opus 5.5 Solves 18.5% of Full Tasks, 56.8% of Steps
On 2026-11-04 Klavis AI released Multi-step SWE (arXiv 2611.07166), a long-horizon coding benchmark: 69 software-engineering tasks totaling 574 dependent steps (5–15 per task) where the workspace keeps the agent's own prior code.
💬 What's your take? Join the discussion in the comments.
Read the original Chinese version
📰 Klavis 发布 Multi-step SWE：69 题共 574 步依赖链，厂商口径 Opus 5.5 整任务仅解出 18.5%、单步 56.8%
Klavis AI 于 2026-11-04 发布长程编码基准 Multi-step SWE（arXiv 2611.07166）：69 个软件工程任务、共 574 个环环相扣步骤（每题 5–15 步），工作区持续保留 Agent 自己写的代码。
💬 你怎么看？欢迎在评论区讨论。
#Multi-step SWE #Klavis #Benchmark
0 Comments