Unverified50% confidenceFactExact time
The core mechanism behind the high token consumption of long-horizon agents is repeated context processing, where the full historical context must be re-fed into the model after each tool call.
1
Sources
50%
Confidence
Long-term
Relevance
10/6/2026
First Seen
Sources
Related Entities
Related Claims
Unverified传统ReAct框架要求模型在每次工具调用后将结果写回上下文再重新读取整个对话历史决定下一步，导致Token消耗和延迟累积78% similarUnverifiedIn each iteration of the loop, an agent needs to repackage the system prompt, conversation history, tool descriptions, and all outputs from previous steps before feeding them into the model, causing token consumption to grow linearly or even super-linearly with the number of steps.76% similarVerifiedThe self-attention mechanism enables models to simultaneously handle long-range text dependencies, breaking through the sequential processing bottleneck of recurrent neural networks (RNNs).71% similarUnverifiedDue to the cyclic reasoning nature of Agent systems, the complete conversation history is resent to the model after each tool call, causing Token consumption to grow super-linearly.70% similarVerifiedLarge model output generation uses an autoregressive approach to produce tokens one by one, requiring recalculation of attention over all preceding context with each token generated.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/979788API
curl https://kongchang.com/api/v1/knowledge/claims/979788MCP
get_claim(id=979788)