Kimi Code has launched the K3.1 Pro series, its latest generation of AI models designed for coding and other long-horizon tasks. The series includes K3.1 Pro and K3.1 Pro Highspeed, both built on the same underlying model but with the Highspeed version delivering approximately two to three times the speed for more responsive interactions.
According to the company, the models offer substantial improvements in coding, 3D creation, financial research, and other agentic capabilities that require sustained reasoning over long contexts. Both models support a 1 million token context window and require thinking mode to be enabled; requests with thinking disabled are rejected.
Kimi Code recommends users start a new session when switching to the K3.1 Pro series, rather than continuing an existing session from older models such as the K3 series, K2.8 Preview, or K2.7 Code Highspeed, to avoid degraded output quality and reduced cache hits. The company also advises using the native Responses, chat completion, or message APIs for the most stable output, as protocol translation through proxy tools may cause model behavior to vary.
The K3.1 Pro series is now available in Kimi Code clients including Desktop, CLI, and the VS Code extension. Users on compatible membership plans can access the models through the model picker or by entering the appropriate model ID for third-party tools. The K3.1 Pro models join the existing lineup of K3, K2.8 Preview, and K2.7 Code Highspeed models.
For users who do not see the new models in their client, the company suggests restarting the application or signing in again.