Kimi K3.1 Pro ranks 4th of 215 models, scrubs cyberattack data before open weights

Moonshot AI released Kimi K3.1 Pro, a post-training upgrade to Kimi K3 with 2.8T parameters and a 1M-token context, ranking fourth among 215 models on the Artificial Analysis Intelligence Index. Before the November 4 open-weight release, Moonshot removed offensive cyberattack training data, dropping its xdev-bench-hard score from 90.5% to 81.0%.

• K3.1 Pro placed 4th among 215 models on the Artificial Analysis Intelligence Index
• It scores 81.0% on xdev-bench-hard versus 90.5% for the unredacted ablation checkpoint
• Kimi Hallu Bench Hard error rate fell from 23.4% to 8.3%; Kimi Code Bench rose from 55.95% to 61.2%
• Open weights arrive November 4; K3 architecture and 1,048,576-token context are unchanged