Moonshot AI's Kimi K3, a 2.8 trillion-parameter model, demonstrates competitive, broadly usable capabilities in coding, reasoning, long-context work, and tool-driven execution, surpassing its predecessor Kimi-K2.7-Code in coding performance. The model's verification workflow, including screenshot comparisons and unit tests, consumes roughly one-third of tokens and half of execution time, contributing to its reliability and operating cost.
Chinese labs are now training trillion-scale systems that compete across multiple frontier-level capabilities, increasing demand for advanced compute.