Unverified50% confidenceFactExact time
LLM推理服务的限流通常不仅基于请求次数，还会基于输入输出的Token总量进行限速
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
美团CatPaw实测：免费GLM-5.2编程工具靠谱吗？
bilibili鲲鹏Talk7/2/2026
Related Claims
Unverified每次批注都要调用 LLM，对高频微调场景，API 成本与响应延迟都是实际约束74% similarUnverifiedAI平台限流通常表现为每分钟请求数或每日Token总量的硬性上限，降级则可能静默将请求路由到参数更小的模型69% similarUnverified对于大多数现代 LLM，推理过程是显存带宽瓶颈而非计算瓶颈68% similarUnverified响应速度越快的产品，单位时间内处理的请求越多，反而越容易触及用量上限67% similarUnverifiedExplicitly specifying which dimensions of information to extract dramatically reduces output uncertainty in LLM prompts.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/113038API
curl https://kongchang.com/api/v1/knowledge/claims/113038MCP
get_claim(id=113038)