Chinese artificial intelligence startup Moonshot AI, best known for its Kimi chatbot, on Tuesday released its next-generation open-source model, Kimi K2.5, which the company said is its most capable model to date.

According to Moonshot AI, Kimi K2.5 achieved the best performance among global open-source models in multiple agent-focused benchmarks, including Humanity’s Last Exam (HLE), BrowseComp, and DeepSearchQA, highlighting its advances in reasoning, task execution and autonomous agent capabilities.

Designed as a general-purpose model, Kimi K2.5 is built on a native multimodal architecture that supports both visual and text inputs. The model integrates visual understanding and reasoning with programming and agent-based functions within a single system, allowing it to handle a wide range of tasks.

Moonshot AI founder and CEO Yang Zhilin said the company rebuilt its reinforcement learning infrastructure and specifically optimized training algorithms to maximize efficiency and performance.

The release comes as Chinese AI developers increasingly embrace open-source strategies in an effort to accelerate adoption and compete with global peers in advanced foundation models.