Primary sourceARC Prize
View sourceEvent ·Multi-source
Anthropic 的 Claude Sonnet 5.5 在 ARC-AGI-3 上拿到 72.4%
ARC Prize 公布了 Anthropic 的 Claude Sonnet 5.5 在 ARC-AGI(Verified)上的成绩:ARC-AGI-3 拿到 72.4%,花费 $8.2K(provider adapter harness),标准 harness 下为 24.3%。ARC-AGI-2 成绩为 83.9%,每任务 $0.55。ARC-AGI-1 达到 96.5%,每任务 $0.14。其中 72.4% 的 ARC-AGI-3 分数被认为是 Sonnet 5.5 最值得注意的表现。
Current read
Anthropic 的 Sonnet 5.5 在 ARC-AGI-3 上拿了 72.4%,这个基准一直很难,数据里还给了不同 harness 的对比和每任务成本,做评测的可以看看。
11 sources79° AI heatLast updated 2026/10/21 19:30:47
Evidence
11 sourcesRelated source 1Artificial Analysis
View sourceRelated source 2rohanpaul_ai
View sourceRelated source 3AWS Machine Learning Blog
View sourceRelated source 4The Information
View sourceRelated source 5kimmonismus
View sourceRelated source 6Andrew Curran
View sourceRelated source 7Ethan Mollick
View sourceRelated source 8Liputan6 Tekno
View sourceRelated source 9Tanishq Abraham (论文推介)
View sourceRelated source 10swyx (AI Engineer)
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.