Unverified50% confidenceBenchmarkExact time
在同一代码仓库、同一模型、同一工作流的对照测试中,通过 Edgee 压缩网关路由的 Codex 输入 token 减少 49.5%
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/18/2026
First Seen
Valid until: 12/17/2026
Sources
Related Entities
Related Claims
Unverified微软研究院推出的LLMLingua等工作证明,通过移除冗余Token可以在保持95%以上任务准确率的前提下压缩40%-60%的输入Token64% similarUnverified当前缀缓存与DFlash2投机解码同时启用时,总吞吐量下降约37%,甚至比完全不开投机解码还要慢64% similarUnverifiedCodeGraph测试显示Token消耗减少约47%63% similarUnverifiedCodeGraph测试显示工具调用次数减少约58%62% similarUnverified混合架构中,仅当规则匹配失败或置信度低于阈值时才调用大模型兜底,通常可减少60%-80%的API调用量62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/932086API
curl https://kongchang.com/api/v1/knowledge/claims/932086MCP
get_claim(id=932086)