On October 24, Moonshot AI proudly announced the launch of Kimi K3.1 Pro along with its high-speed iteration, K3.1 Pro Highspeed. The latter variant impressively operates at a speed 2 to 3 times swifter than its standard counterpart. Both innovative models boast a substantial context window capacity, extending up to one million tokens, thereby rendering them apt for a diverse array of domains including comprehensive software project development, game creation, 3D modeling, and financial scrutiny. According to official statistics, the new models attained a commendable 65.7% in programming assessments, surpassing GPT-6 Astra's 58.2%; and garnered a 92.7% score in 3D modeling evaluations, approaching the proficiency of premier closed-source models. In the realm of financial analysis, these models excel in processing financial data, cross-referencing documents, and executing investment analyses, with their performance in corporate operation tests approximately quadrupling that of their predecessors (about 4.2 times). Notably, the hallucination rate per statement in specialized tests has plummeted from 23.4% in the previous generation to a mere 8.3%. Presently, both models are accessible on Kimi.ai, Kimi Work, Kimi Code, and through API, with the complete model weights slated for release by November 4.