Kimi K1.5 is Moonshot AI’s first reasoning-focused model, released in January 2025 as the company’s answer to OpenAI’s o1. It is trained with reinforcement learning on long chains of thought and can reason jointly over text and images, which let Moonshot claim results on par with o1 on math, coding, and multimodal benchmarks like MathVista, AIME, and Codeforces. The model supports a 128,000-token context window and includes a “long2short” technique that transfers reasoning gains from long chains of thought into shorter, faster responses for everyday queries. Moonshot has not disclosed the parameter count. K1.5 was Moonshot’s proof that a Chinese lab could compete at the reasoning frontier, and it set the stage for the open-weight Kimi K2 line that followed a few months later.