Unverified50% confidenceBenchmarkExact time
在 40 个任务的学术基准 Cybench 上使用 Opus 4.8（去掉提示与题解），T3MP3ST 取得 23/40 = 58% 的单次无提示 pass@1
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/6/2026
First Seen
Valid until: 10/4/2026
Sources
T3MP3ST：将编程Agent武装为自主渗透红队的开源框架
twitterelder_plinius7/5/2026
Related Claims
UnverifiedClaude Opus 4.8 completed identical tasks with 15% fewer steps and 35% fewer output tokens compared to its predecessor66% similarUnverifiedOpus 4.7每次解题的中位输出token数为60,000，而GPT 5.5仅需16,00065% similarUnverifiedClaude Opus 4.7中间检查点通过率达57%，最终准确率却只有39.8%65% similarUnverifiedQwen3.6 35B在SWE Bench Verified上的成绩距离Claude Opus 4.6仅差1.6个百分点64% similarUnverified在 XBEN 白盒模式下 T3MP3ST 的 pass@1 达到 98.7%，最差单轮为 98.1%（102/104）63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/114276API
curl https://kongchang.com/api/v1/knowledge/claims/114276MCP
get_claim(id=114276)