Moonshot's Kimi K2.8 Preview Rolls Out on Kimi Code, Performance Close to K3

Chinese AI company Moonshot says its Kimi K2.8 Preview model is now fully available on Kimi Code. The Model ID stays the same as kimi-for-coding, so clients and third-party tools can use it without changing any settings.

The company says K2.8 Preview performs close to K3 overall, with better thinking efficiency. Its coding and agent abilities are improved across the board, and thinking is much better than K2.7 Code. All membership tiers now get a context window of up to 1M.

The model supports thinking effort levels, with low, high, and max options, defaulting to max, matching K3's thinking levels. When thinking is turned off, requests from both the K3 series and K2.8 Preview are routed to K2.8 Preview without thinking.