Unverified50% confidenceFactExact time
Industry estimates suggest that implicit tokens (system prompts, RAG retrieval context, function descriptions, and conversation history) typically account for 60%–80% of total consumption per request.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/3/2026
First Seen
Valid until: 12/2/2026
Sources
Related Entities
Related Claims
Unverified在设计不良的多智能体系统中，协调 Token（Agent 间传递的指令、确认、状态摘要）可占总消耗的40%-60%66% similarUnverifiedIn actual deployments, enterprises commonly find that 70–80% of daily queries can be handled by smaller models.64% similarUnverified研究显示复杂代理任务的token消耗往往是等效单次对话的20-100倍63% similarUnverified超低输入价格对RAG场景有利，因为典型RAG查询可能注入5000-20000个token的检索结果而输出仅有200-500个token63% similarUnverifiedMeta、NVIDIA等公司的研究表明RAG在知识密集型任务上的准确率比纯参数记忆提高20%-40%62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/847624API
curl https://kongchang.com/api/v1/knowledge/claims/847624MCP
get_claim(id=847624)