Unverified50% confidenceTradeoffExact time
智能体模式下Token消耗呈非线性增长，因每次工具调用结果追加到上下文中，多步骤任务的Token消耗可达O(n²)级别，使低成本模型对智能体场景至关重要
1
Sources
50%
Confidence
Long-term
Relevance
8/18/2026
First Seen
Sources
Related Claims
Unverified在智能体工作流中，历史交互作为输入Token累积，导致总Token消耗随任务轮次呈近似平方级增长76% similarUnverifiedAgentic systems often run continuously every minute via scheduled tasks, and combined with multiple rounds of model calls, tool call results being fed back in, and reflection steps, token costs escalate exponentially.74% similarUnverifiedMulti-agent parallel architectures explore multiple solution paths simultaneously, trading higher compute for lower error rates and shorter wall-clock time, but at the cost of a multiplicative increase in token consumption.74% similarUnverifiedThe high consumption triggered by the combination of Extra High reasoning and multi-tool agent tasks is expected behavior, not a system bug.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/771252API
curl https://kongchang.com/api/v1/knowledge/claims/771252MCP
get_claim(id=771252)