Unverified50% confidenceFactExact time
A query that takes 2 seconds in standard mode may become 25 seconds in deep reasoning mode, while consuming 5 to 10 times the tokens.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/15/2026
First Seen
Valid until: 1/13/2027
Sources
Related Entities
Related Claims
Unverified官方自研推理运行时在处理 1M token 文档时约 3 秒给出决策，完整读取一遍约需 111 秒76% similarUnverifiedIn actual testing, a single classification by the Dify question classifier takes approximately 11 seconds, because execution speed depends on the large model's response.75% similarUnverifiedOfficial data shows that on an entry-level GPU, querying 10 questions together takes a total of 72 milliseconds, averaging approximately 7 milliseconds per question.74% similarUnverified对于BERT-base模型推理会话初始化通常耗时数百毫秒，重复执行会使延迟增加1-2个数量级73% similarUnverifiedAfter modern tokenizers are highly optimized, the time for a single tokenization operation is typically on the order of 10–50 ms73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1035601API
curl https://kongchang.com/api/v1/knowledge/claims/1035601MCP
get_claim(id=1035601)