Primary sourceFireworks AI
View sourceEvent ·Multi-source
DeepSWE测试显示DeepSeek V4.1-Flash输入token占比远高于输出
DeepSWE测试对比Astra和DeepSeek V4.1-Flash,输入token数量是输出的174倍。99.6%的请求是缓存命中,占60%的费用。最终任务质量相同,但成本从$6.52降至$0.43。
Current read
DeepSWE测试发现DeepSeek V4.1-Flash的输入token消耗远高于输出,缓存命中率高,能大幅降低成本。
11 sources58° AI heatLast updated 2026/9/15 19:17:43
Evidence
11 sourcesRelated source 1orange.ai
View sourceRelated source 2歸藏(guizang.ai)
View sourceRelated source 3DeepLearning.AI
View sourceRelated source 4lmarena.ai
View sourceRelated source 5AI Will
View sourceRelated source 6OpenAI Blog
View sourceRelated source 7量子位
View sourceRelated source 8elvis
View sourceRelated source 9Gary Marcus
View sourceRelated source 10a16z
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.