Expired60% confidenceBenchmarkExact time
When Sonnet's context exceeds the 50,000 to 100,000 token range, its accuracy in locating repeated words within the context window drops by nearly 50%.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/9/2026
First Seen
Valid until: 10/7/2026(expired)
Sources
Related Claims
Unverified智能体能力的提升伴随Token消耗的指数级增长，Sonnet 5在最大努力模式下Token消耗量堪比OPUS但效果更差63% similarUnverifiedThe article recommends using Sonnet 70% of the time for daily coding and Opus 30% of the time for complex reasoning.62% similarUnverifiedOn GDPval, the Max Effort cost of Sonnet 5 is approximately 6 times that of Low Effort.62% similarUnverifiedExtending flagship-level Agent capabilities to the mid-tier Sonnet model means similar tasks can be completed at approximately one-tenth the unit cost.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/400640API
curl https://kongchang.com/api/v1/knowledge/claims/400640MCP
get_claim(id=400640)