Kimi K3.1 Pro brings a new long-context model and a refreshed coding workspace

Moonshot has rolled out K3.1 Pro and Highspeed, while Kimi Code adds workspace trust, model-switch choices and session reload.

Moonshot's latest release combines a model upgrade with changes to the desktop and terminal environments developers use every day. Kimi K3.1 Pro began rolling out on 23 October, alongside a Highspeed variant. The company's technical announcement lists availability across Kimi, Kimi Work, Kimi Code and its API, with model weights scheduled for 4 November.

A larger working context for extended projects

The technical blog describes a 3T-class model with a one-million-token context window and post-training focused on coding, three-dimensional creation and financial research. Moonshot reports a 65.7% result on Terminal-Bench 4.0 and 61.2% on its internal Kimi Code Bench 3.0 Long-horizon evaluation. Highspeed uses the same model quality and is advertised at roughly two to three times the speed of the standard Pro variant.

The launch also reaches creative work. Kimi's 3D Make feature lets users preview and edit models online and export them for printing or other media. The model announcement describes scripting workflows involving Blender and CAD, connecting language instructions with editable geometry rather than a finished picture alone.

Changes inside Kimi Code

Kimi Code's 21 October desktop release introduces a trust question when a folder is opened. Model switching offers three choices: start a new session, compact the context and continue, or continue directly. A session reload now applies changes to AGENTS.md, skills and external configuration without restarting the application. The release also addresses streaming-related stutter and device heating in long conversations.

The Code documentation gives practical setup details for K3.1 Pro. Thinking remains enabled, with low, high and max levels. A new session is recommended when moving from an older model family. Available context and Highspeed access depend on the selected membership tier.

Together, the changes make this a useful release to assess as a complete working environment: model selection, project context, configuration loading and review remain part of the same developer journey. A team can keep a representative repository, compare the resulting patch and application behaviour, and record which model and session settings produced the result. That makes the upgrade repeatable across colleagues rather than dependent on one successful chat.

Sources

- [K3.1 Pro Series is live](https://www.kimi.com/code/docs/en/kimi-code/whats-new.html) — Moonshot AI · 2026-10-23 - [Introducing Kimi K3.1 Pro](https://www.kimi.ai/blog/kimi-k3-1-pro) — Moonshot AI · 2026-10-23