Kimi K3.1 Pro nears Claude Opus 5.5, weights due Nov. 4

The frontier just got a near neighbor you'll be able to download.

Moonshot AI released Kimi K3.1 Pro, which it calls its most capable 3T-class open-weight model to date. It has a 1M-token context window. Moonshot says full model weights will be released by Nov. 4, 2026. Until then it's an API and app model, not yet an open-weight model.

The numbers Moonshot published

K3.1 Pro is a post-trained version of Kimi K3. A second version, K3.1 Pro Highspeed, runs two to three times faster, Moonshot says.

On Terminal-Bench 4.0, a test of agentic work in a command line, Moonshot reports:

65.7% for K3.1 Pro, against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra.

Those are Moonshot's own runs, at maximum reasoning effort. Treat them as a claim.

There is an outside read. Simon Willison reports that Artificial Analysis scored K3.1 Pro at 55, behind only Claude Opus 5.5 at 58 and ahead of Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon at 53 each. He calls it "the first time a (soon-to-be) open weight model has scored this highly against the frontier."

Moonshot also says it cut hallucination sharply. On its own hard factual benchmark, the claim-level hallucination rate fell from 23.4% for K3 to 8.3% for K3.1 Pro.

One more detail is worth noticing. On Moonshot's internal cybersecurity test, K3.1 Pro scored 81.0, below the 90.5 of a Kimi model without its cyber risk reduction. Moonshot shipped the weaker one on purpose, by its own chart.

Why the weights are the story

Yes, another model launch with a vendor benchmark table. The reason to care is the license file that isn't here yet.

A model this close to the top that you can run yourself changes three things for you. You can host it where your data already lives. You can fine-tune it. And every closed lab's pricing now has a floor set by whatever it costs you to run this.

Moonshot hasn't said what license the weights will carry. That decides whether any of this applies to your business.

Price and where to try it today

Willison reports API pricing of $2.00 per million input tokens and $10.00 per million output tokens. He says that output price is half of Claude Opus 5.5's. Highspeed costs twice as much, he reports.

Moonshot lists these places to use it now: Kimi.ai and the Kimi mobile apps, the Kimi Work desktop app (version 3.1.0 or later), Kimi Code through the /model command, and the Kimi API as kimi-k3.1-pro and kimi-k3.1-pro-highspeed.

Two details for teams. Kimi Enterprise admins have to switch K3.1 Pro on, since it's off by default at launch. And Moonshot offers zero data retention to API customers.

If you work somewhere with rules about Chinese AI vendors, check them first. TopFive covered one such rule when DHS banned DeepSeek from federal agencies.

Nov. 4 is the date to circle

Three things to check when the weights arrive: the license, the hardware it takes to run a 3T-class model, and whether independent tests match Moonshot's table.

Run your own evals this week. Decide on Nov. 4.