Expired50% confidenceBenchmarkExact time

Claude Opus 4.8 achieved a completion rate of only 20.6% on OSWorld 2.0 with a 500-step budget, a partial score of 54.8%, and an average cost of approximately $72.4 per task.

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

7/9/2026

First Seen

Valid until: 10/7/2026(expired)

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

UnverifiedIn the e-commerce website test, Opus 4.8's estimated cost was $21.41 while Fable 5's was $36.8472% similar UnverifiedGPT-5.5 achieved a completion rate of 13.0% on OSWorld 2.0, with a cost of approximately $25.5 and the highest token efficiency.69% similar UnverifiedUsing the flagship model Claude Opus to complete a task involving 20 tool calls with an average context of 10,000 tokens per round, the input cost alone could exceed $3.68% similar UnverifiedPathway模型论文称在ARC-AGI等基准上达到29.5%成绩，单题计算成本仅为0.0007美元68% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/338174

API

curl https://kongchang.com/api/v1/knowledge/claims/338174

MCP

get_claim(id=338174)