Moonshot AI Releases Kimi K2.8 Preview Model on Kimi Code
Moonshot AI has officially launched the Kimi K2.8 Preview model across Kimi Code, bringing overall performance close to its flagship K3 model with higher thinking efficiency. The update upgrades coding and AI agent capabilities while opening a 1M token context window to all member tiers. This release provides developers with near-flagship AI reasoning and long-context capabilities without requiring any configuration changes in existing client tools or workflows. It underscores the trend of bringing high-level reasoning controls and large context windows to mainstream AI coding tools. The model retains the existing `kimi-for-coding` ID and supports three adjustable thinking effort levels: low, high, and max (defaulting to max). When the thinking feature is turned off, requests for both K3 series and K2.8 Preview route to the non-thinking K2.8 Preview variant.
## BACKGROUND
Moonshot AI is an AI company known for its Kimi series of large language models and Kimi Code development tools, which emphasize long-context processing and AI coding assistance. Reasoning LLMs increasingly incorporate configurable thinking effort levels, allowing developers to balance output accuracy, latency, and resource consumption based on task complexity.