Kimi K3.1 Pro, released October 24, 2026, by Moonshot AI as a post-trained upgrade to the July Kimi K3, immediately set the leading HLE score at 70.0% with tools on the 2,500-question expert benchmark, ahead of Claude Opus 5.5 at 67.7%. The 2.8T-parameter MoE model with 1M context and multimodal support delivers gains in agentic coding, long-horizon tasks, and knowledge reasoning through scaled reinforcement learning. Full weights are scheduled for release by November 4, enabling independent verification and potential fine-tuning. Rival labs continue iterative releases, but current gaps and the benchmark’s resistance to saturation suggest limited upside for new Kimi variants before year-end absent major undisclosed advances.