Unverified60% confidenceFactTime unknown
部分AI Agent系统在SWE-bench上的解题率已突破50%
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026
Sources
SWE-bench官方博客上线：AI编程评测标准进入新阶段
twitterSWEbench
Related Entities
Related Claims
Unverified目前顶尖AI Agent（如Devin、SWE-agent）在SWE-bench上的解决率约为40-55%80% similarUnverifiedAI Agent同一个任务在不同上下文长度下，单次Token消耗能飘移5到50倍77% similarUnverified研究表明给 AI 提供足够的上下文约束能将输出代码的可用率提升 50% 以上73% similarUnverifiedAccording to Pragmatic Engineer survey data, 55% of code is now generated by AI Agents71% similarUnverifiedAI Agent工作流的算力消耗可能是单轮对话的50到100倍以上71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/26926API
curl https://kongchang.com/api/v1/knowledge/claims/26926MCP
get_claim(id=26926)