Unverified50% confidenceBenchmarkExact time
In the P2NPCBench evaluation, 12 proprietary and open-source models achieved an average pass@1 of 35.4% on NPC problems and 63.8% on P twins, a gap of nearly 28 percentage points.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/22/2026
First Seen
Valid until: 1/20/2027
Sources
Related Entities
Related Claims
Unverified二项分布的方差为Np(1-p)，在p=0.5时达到最大值64% similarUnverified测试数据表明RNN头与马尔可夫头的草稿通过率差距仅在1-2%以内，而部署成本差距可达数倍63% similarUnverifiedIn the specific experimental case, the overlapping-anchor scheme (n=912) showed the model at 55.2% versus the majority class at 51.1%, a net advantage of +4.1pp, about 2.4 standard errors; the non-overlapping-anchor scheme (n=163) showed 59.5% versus 54.0%, a net advantage of +5.5pp, about 1.4 standard errors.61% similarUnverified若单一模型漏检率为 p，两个训练背景不同的模型同时漏检同一错误的概率近似降至 p²，20%漏检率理论上可降至 4%61% similarUnverifiedThe model's recall for high/VP attachment (N1) is as high as 99.71%, while its recall for low/NP/embedded attachment (N2) is only 89.26%.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084108API
curl https://kongchang.com/api/v1/knowledge/claims/1084108MCP
get_claim(id=1084108)