Kimi K2.7 Code: China's Open Coding AI Advances

Moonshot AI has launched a new coding model called Kimi K2.7 Code. This AI program helps write computer code faster and can handle complex, multi-step coding tasks independently. Here's the thing: The company reports that Kimi K2.7 Code scores 21.8% higher than its previous model, Kimi K2.6, on a key coding test. This model is "agentic," meaning it can plan and complete long tasks step-by-step. It's also "open weights," allowing others to download and run the model themselves under a Modified MIT license. What's interesting is how it's built. Kimi K2.7 Code is a Mixture-of-Experts model with one trillion total parameters, though only 32 billion activate for each piece of work. This makes it powerful yet efficient. It has 384 experts and 61 layers, and it includes a vision encoder, MoonViT, allowing it to process images and video. The model can hold 256K tokens of text in its context window, which is roughly the size of a small book. The bottom line: This development shows progress in China's open coding models and could mean more accessible and powerful AI tools for software development.