Unverified50% confidenceTradeoffExact time
In chunk granularity selection, chunks of 128–512 tokens are generally suitable for precise retrieval, chunks of 512–2,048 tokens better preserve context, and retaining 10–20% overlapping text between adjacent chunks can mitigate information loss.
1
Sources
50%
Confidence
Long-term
Relevance
9/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen splitting text by token count, a chunk size of 200–500 tokens is typically set.74% similarUnverified文档切分中块太大则检索精度下降，块太小则丢失上下文语义，设置50-100 token的重叠能避免关键信息被切断74% similarUnverifiedIn traditional RAG document chunking, documents are split into fixed-length segments (e.g., 256 characters) with some overlap between segments to maintain semantic coherence65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/849933API
curl https://kongchang.com/api/v1/knowledge/claims/849933MCP
get_claim(id=849933)