Expired50% confidenceTradeoffExact time
The improvement in agentic capabilities is accompanied by exponential growth in token consumption, and Sonnet 5's token consumption in maximum-effort mode is comparable to OPUS but with worse results.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/22/2026
First Seen
Valid until: 10/20/2026(expired)
Sources
Related Claims
UnverifiedOn GDPval, the Max Effort cost of Sonnet 5 is approximately 6 times that of Low Effort.72% similarUnverifiedEarly testing of Claude Sonnet 5 shows that its improved reasoning capability comes at the cost of consuming approximately 30% more tokens for the same prompt.65% similarUnverifiedWhen 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%.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/589918API
curl https://kongchang.com/api/v1/knowledge/claims/589918MCP
get_claim(id=589918)