Kimi K3.1 Pro Launches With a 1 Million-Token Context Window

Kimi K3.1 Pro is now available, with a one-million-token context window for working with long documents and large code projects. Moonshot AI released the model on October 23, 2026. You can access it through Kimi’s apps, Kimi Work, Kimi Code and its API. A faster Highspeed version is also available.

A context window sets how much material a model can take into account in one request. However, a larger window does not guarantee correct answers. For developers, the appeal is practical: you may be able to supply more of a codebase or document collection without splitting it into many small requests.

What Kimi K3.1 Pro offers today

In its technical announcement, Moonshot describes the release as an improved version of Kimi K3. The company says it has increased training after the initial model training stage. It targets coding, financial research, 3D creation and tasks that require several steps.

Separately, Vercel confirmed availability on October 23. Developers can call both versions through its AI Gateway. This provides another access route, although it does not independently prove Moonshot’s performance claims.

Moonshot reports a 65.7% score on Terminal-Bench 4.0, a test for coding agents. Treat that number as a company-reported result. Your own project may involve different tools, instructions and constraints. Therefore, a small trial on your actual workload will tell you more than a single benchmark score.

What still needs checking

Full model weights are due by November 4, according to Moonshot. Until then, this is a hosted-service launch. People hoping to download and run the model themselves still need to wait. A promised release date is a future milestone, rather than something you can use today.

The company also says it removed cyberattack-related training material to reduce misuse risk. It reports that the final model remains stronger than K3 on an internal vulnerability test. However, it scores below a version without those cuts. These are Moonshot’s own tests, rather than an outside safety audit.

The Next Web’s independent reporting also notes that the benchmark figures lack independent verification. As a result, neither the scores nor the training changes establish that the model is safe for every use.

For now, Kimi K3.1 Pro gives developers another hosted option for long-context work. Start with a small task, check the output and review the service’s data policies before sending sensitive material. For other recent developments, browse Dumb IT Dude’s News archive.

Archive photograph of server racks in a data center, photographed by Carl Lender in 2015. Photo: Carl Lender from Sunrise, USA / CC BY 2.0. Resized/display crop; original image license applies.