Kimi Code rolls out K3.1 Pro and K3.1 Pro Highspeed with 1M context

Kimi Code has fully rolled out K3.1 Pro and K3.1 Pro Highspeed, the first models in the K3.1 series and the company's most capable 3T-class offerings to date.
K3.1 Pro supports a 1M context window and shows substantial improvements in coding, 3D creation, and financial research, while K3.1 Pro Highspeed delivers identical capabilities at approximately 2-3x the speed.
Both models enforce thinking mode exclusively, rejecting any requests with thinking disabled, and require users to start new sessions to avoid quality degradation and cache invalidation.
The API supports OpenAI and Anthropic protocols, with specific base URLs provided for China and overseas regions to facilitate integration with third-party tools like Claude Code and Codex.

Model Specifications and Availability

K3.1 Pro and Highspeed require Plus-tier plans for access, with 1M context reserved for Pro-tier and above members.
K3.1 Pro Highspeed consumes approximately 2x the quota of the standard version, whereas the previous K2.7 Code Highspeed runs at 5-6x speed with 3x quota usage.
Multimodal input for the new models includes both image and video, matching the capabilities of the previous K3 and K2.8 Preview models.

Users must use specific model IDs (k3.1-pro, k3.1-pro-highspeed) in third-party tools, as mistyped IDs silently fall back to K2.8 Preview without thinking.
Thinking effort levels (low, high, max) are supported, with max as the default for K3.1 Pro models; changing effort levels invalidates the context cache and increases token usage.