Kimi K2.7 Code is a coding-focused agentic model in Moonshot AI's Kimi K2 family, built to carry out end-to-end programming tasks reliably over very long contexts. It uses a native multimodal mixture-of-experts architecture that accepts text, image, and video inputs while producing text output, and it operates in a persistent thinking mode that preserves full reasoning content across multi-turn conversations. Under the hood it scales to roughly one trillion total parameters while activating about 32 billion per token, spread across 384 experts with 61 layers, giving it the headroom to handle long-horizon code generation, agentic task decomposition, and tool-driven workflows at a 256K-token context window.

The model is an evolution of Kimi K2.6, refining that base for software engineering with substantial gains on real-world long-horizon coding tasks and stronger instruction compliance in extended contexts. External benchmark evaluations show it reduces overthinking tendencies by around 30 percent on average and cuts thinking-token usage by roughly 30 percent compared with its predecessor, translating to more efficient end-to-end task completion. A high-speed variant pushes output toward 180 tokens per second and up to 260 in short contexts, and open weights make it straightforward to fine-tune with techniques such as LoRA. In practice it fits teams that need a transparent, agentic coding partner for large repositories, multi-step refactors, and tool-using assistants that benefit from always-on reasoning.