Verified65% confidenceFactExact time
OSWorld 2.0 contains 108 long-workflow real computer tasks, with a median completion time of 1.6 hours for skilled humans, approximately 48 times that of version 1.0 (2 minutes).
3
Sources
65%
Confidence
Long-term
Relevance
7/9/2026
First Seen
Sources
OSWorld 2.0测评：最强AI Agent完成率仅20%，长流程是致命短板
bilibili熊二等兵7/1/2026
Related Claims
UnverifiedAndroid Studio 启动通常需要数十秒，占用内存动辄超过 2GB65% similarUnverifiedCompleting the 30-second version involved 14 sub-agents working in coordination, with the entire process involving 779 tool calls (161 in the main conversation and 618 in sub-agents), taking approximately 2.4 hours in total.63% similarUnverified微软研究院2022年发布的研究显示，开发者平均每天只有约2-4小时处于真正的深度编码状态，这段时间产出的代码量和质量往往占全天工作成果的70%以上63% similarUnverifiedThe tropical island project used multiple sub-agents, with the API running for approximately 2 hours and 7 minutes, equivalent to roughly 8 hours in real time.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/338169API
curl https://kongchang.com/api/v1/knowledge/claims/338169MCP
get_claim(id=338169)