AI 导读
每任务耗时包含实测搜索时间和模型时间。Perplexity Search 处于中游：每次搜索查询约 1.1s，medium 和 high 变体每任务约 27 至 29s，而榜单整体范围约为 19 至 62s。low 变体约需 36s：每次查询速度相同，但搜索次数更多。更详细的载荷导致更少的智能体动作：在 BrowseComp 上，high 每任务 13.1 次搜索，而 low 为 18.0 次。
正文
Time per task includes both measured search time and model time. Perplexity Search sits mid-pack: ~1.1s per search query, and roughly 27 to 29s per task for the medium and high variants against a board range of about 19 to 62s. The low variant takes about 36s: same per-query speed, more searches. More detailed payloads result in fewer agent actions: 13.1 searches per task for high vs 18.0 for low on BrowseComp.
来源：@ArtificialAnlys · x.com