On October 28, Qianlong and Sina reported that Moonshot AI has released a new-generation open-source model, Kimi K3.1 Pro — the world's largest-parameter open-source model, with comprehensive intelligence approaching top closed models at about one-fifth the price of some rivals. The iteration reinforces a domestic open-source acceleration curve rather than a single company's lone push.
K3.1 Pro strengthened code writing, 3D creation and financial-research capabilities, and can complete longer multi-step complex tasks. In an internal test, researchers had the model build a simulated human-walking system; it continuously analyzed results, located problems and iteratively optimized the controller, advancing a simulated human from falling after a few steps to alternating steps and walking continuously for over ten seconds.
The report ties Kimi's direction to Beijing's systemic AI layout — the city's "AI soft eight measures," the July "measures to accelerate agent-led development," and support for online learning, continuous learning and autonomous evolution. Alongside DeepSeek, Zhipu GLM-5.3-Flash and others, domestic open models are converging into an upward force.
As one observer put it, if the closed-source race is about "who holds the ticket," the open-source route answers "how wide the road can be paved."
Source: Qianlong / Sina
