Moonshot AI's Kimi Code documentation lists four coding-model IDs — k3, k3-256k, kimi-for-coding and kimi-for-coding-highspeed — spanning K3, K2.8 Preview and K2.7 Code HighSpeed¹.

The docs describe K3 as a flagship coding model with 2.8 trillion parameters and a 1,048,576-token context window for higher-tier members, while k3-256k is a 262,144-token variant that uses less quota¹.

kimi-for-coding is identified as K2.8 Preview, which the company says is fully rolled out in Kimi Code with performance close to K3, low/high/max thinking effort and up to 1 million tokens of context; kimi-for-coding-highspeed is K2.7 Code HighSpeed, with the same coding ability as that line and about five to six times faster output at three times quota usage¹.

Availability is gated by plan: on new plans, K3 and K2.8 Preview start at Plus, 1 million-token K3 context at Pro and above, and HighSpeed at Pro and above; the docs say requests beyond a plan return HTTP 401¹.

K3 and K2.8 Preview accept image and video input; k3-256k is image-only¹. Kimi Code's API is documented with OpenAI- and Anthropic-compatible base URLs for China (api.kimi.com) and overseas (api.kimi.ai)¹.

The company recommends starting a new session when switching model IDs because cache from a prior model does not hit, which can raise usage¹. HighSpeed speeds only model output, not tool calls or scripts, and a mistyped HighSpeed ID silently falls back to standard kimi-for-coding¹.

Sources

1. kimi.comModel Configuration | Kimi Code DocsOct 23, 12:59 PM

Read the original at kimi.com ↗