Moonshot AI released Kimi K3.1 Pro on October 23, a 2.8T-total/104B-active MoE large language model with 1M context and multimodal support, as a post-trained upgrade to July’s Kimi K3. Self-reported results show 70% on HLE with tools and 65.7% on Terminal-Bench 4.0, placing it near Claude Opus 5.5, while independent Artificial Analysis scored it 52.5% text-only. The open-weights release due by November 4 will enable the first standardized run on the 1,000-question HLE-Diamond subset, where frontier models currently sit in the mid-50s without tools. Traders are weighing these post-training gains in expert reasoning against the historical 22% predecessor score and evaluation discrepancies, with upcoming independent harness results likely to drive odds on debut thresholds.