Unverified50% confidenceFactExact time
One trace shows that a single GraphRAG global query requires 10+ LLM calls and about 200,000 tokens, while a local query takes about 11,500 tokens.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
Unverified一次典型的RAG调用可能检索5-10个文档块，每块500-1000个Token，输入Token总量可达5000-10000个，而生成回答通常只有200-500个Token74% similarUnverifiedIn a typical scenario of 20 rounds of interaction, if the system prompt occupies 15,000 tokens and each round produces an average of 2,000 tokens, by the 20th round the input token count for a single request balloons to 55,000 tokens.65% similarUnverifiedBefore adding the GIN index and the generated tsvector, searching across 10,000 translations took about 800ms; after adding the index, even when the data volume grew to 100,000 translations, the same query dropped to about 20-40ms.65% similarUnverified在一个典型项目查询中发送了45,000 tokens的上下文，但真正有用的只有约5,000 tokens62% similarUnverifiedAll six compared models claim a context window of about 1 million tokens.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1063913API
curl https://kongchang.com/api/v1/knowledge/claims/1063913MCP
get_claim(id=1063913)