Moonshot releases Kimi K3.1 Pro after stripping cyber data from training

Moonshot AI released Kimi K3.1 Pro on October 24, a roughly 3-trillion-parameter open-weight model with a 1-million-token context; weights are promised by November 4. The lab deliberately removed cyber-offense data from pretraining and RL, cutting its score on in-house exploit tasks to 81.0% from 90.5%. On Terminal-Bench 4.0 it scores 65.7% versus 66.4% for Claude Opus 5.5.

• Kimi K3.1 Pro: ~3T parameters, 1M-token context, weights due November 4
• Terminal-Bench 4.0: 65.7% versus 66.4% for Claude Opus 5.5
• Without cyber data the model solves 81.0% of exploit tasks versus 90.5%
• Model iterated a simulated humanoid walk to 11 seconds, fixing 36 errors