Unverified50% confidenceBenchmarkExact time
在基准测试中,从x-high到Max模式,token使用量暴涨1500%(约15倍)
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/1/2026
First Seen
Valid until: 12/30/2026
Sources
Related Entities
Related Claims
UnverifiedMax级别相比extra high最多多消耗两倍的token,但在各类基准测试中仅带来4%至10%的性能提升72% similarUnverifiedXHigh 档位推理 Token 消耗可能达到 High 档位的 3-5 倍,但准确率边际提升通常仅有 5-15 个百分点71% similarUnverified在全文筛选阶段,研究人员编写的理由说明能使模型性能提升约15%68% similarUnverified多角色流水线(Self-Refine)的输出质量通常比单次LLM调用提升15%-30%67% similarUnverified在Skatebench测试中,从x-high档到max档,每题推理token从约330暴涨到近5000(超过10倍),得分只提升1%67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/964511API
curl https://kongchang.com/api/v1/knowledge/claims/964511MCP
get_claim(id=964511)