Unverified50% confidenceBenchmarkExact time
On Terminal-Bench 4.0, Carbon's successful cheating rate dropped to 0.0% and cheating attempts dropped to 0.4%; in internal real coding environments, high-severity misalignment alerts fell by nearly 80%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/28/2026
First Seen
Valid until: 1/26/2027
Sources
Related Entities
Related Claims
UnverifiedAccording to hands-on testing by Bilibili UP host "1024工程笔记," integrating Ponytail reduced code volume by 54%, lowered token consumption by approximately 20%, and kept security-related check items at 100% passing.65% similarVerifiedWhen a single code review exceeds 400 lines of code, the reviewer's defect detection rate drops sharply.64% similarUnverified在一个396个文件的大型混合职责项目上测试时，召回率几乎跌到零63% similarUnverifiedIn a real-world case, a coding agent's success rate dropped from 92% to 88% without any changes, which was a false positive rather than a true capability regression.62% similarUnverifiedIn an A/B test on 50% of traffic, Coinbase's agent reduced fraudulent transactions by 30% and fraudulent amounts by 22%, with the fraud losses saved being about 3 times the inference cost.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1108477API
curl https://kongchang.com/api/v1/knowledge/claims/1108477MCP
get_claim(id=1108477)