Unverified50% confidenceFactExact time
Skill描述通常只有两句话，是每次模型调用都要支付的固定成本，约100-200 tokens
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
Unverified在代理式系统中，一个典型的智能体任务经过8轮迭代后总Token消耗可能达到3-5万Token，成本是单次调用的10-20倍70% similarUnverifiedInference cost per million tokens ranges from a few dollars to tens of dollars, depending on model size and optimization.70% similarUnverifiedAPI中转模式下，每百万Token的成本通常在几美元到十几美元之间，中等强度编程辅助使用月均成本可控制在5-20美元70% similarUnverified单次提示词优化消耗约1000-2000个输入token和500-1500个输出token，额外成本大约在0.02-0.05美元之间67% similarVerified大模型API通常按Token数量计费，大约每750个英文单词或400个中文字符对应1000个Token67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/563122API
curl https://kongchang.com/api/v1/knowledge/claims/563122MCP
get_claim(id=563122)