Kimi K3.1 Pro is out. Moonshot has launched it across Kimi’s website and mobile apps, Kimi Work, Kimi Code and the API. It comes with a one-million-token context window and a Highspeed version which the company says runs two to three times faster than standard Pro. 1
That was a quick turnaround from this morning’s disappearing tech-blog reference. The report is now public, and our screenshot shows K3.1 Pro High selected in Kimi’s model picker. Earlier in the week, we had also found the K3.1 switch rule in Kimi Code. 2
The coding numbers are worth a look
The result that caught our attention is 65.7% on Terminal-Bench 4.0. Kimi’s chart puts that just behind Claude Opus 5.5 at 66.4% and ahead of GPT-6 Astra at 58.2%. On SWE-Marathon, it reports 51.7%, above Opus’s 46.3% and Astra’s 42.5%. 1 3
The results vary quite a bit by task. On Terminal-Bench Science 0.1, K3.1 Pro scores 38.1%, while Astra reaches 68.1% and Opus 63.3%. 1
Kimi says its results use maximum reasoning effort unless otherwise specified. Its coding evaluations use the Claude Code harness, while comparison scores come from a mixture of model reports and public leaderboards. We have not independently rerun these tests. 1
More attention to 3D work
K3.1 Pro is a further post-trained version of K3. Alongside coding, Moonshot is pushing 3D creation and longer research tasks. Its new 3D Make feature lets people preview and edit models in Kimi, download them for 3D printing or export videos and Live Photos. On BenchCAD with tools, it reports 92.7%, compared with Opus’s 96.2% and Astra’s 95.9%. 1
That fits what we have enjoyed in our own testing. Before the launch, we shared a 3D train built from a single prompt, with a speed slider, exploded views, wireframe mode and live stats. The interfaces have felt quicker to produce than with K3, with a stronger eye for presentation. 4
Available now; weights follow by 4 November
In Kimi Code, select K3.1 Pro through /model. API users get kimi-k3.1-pro and kimi-k3.1-pro-highspeed. Kimi Work needs version 3.1.0 or later, and enterprise administrators need to enable access, which is off by default at launch. 1
The hosted release is live now. Full model weights are promised by 4 November 2026. Moonshot calls this the first model in its K3.1 series, so today’s Pro release may not be the whole line-up. 1
Sources & references
- Kimi K3.1 Pro: Advancing Open Intelligence ↗
23 October 2026. Official release, availability, Highspeed claim, million-token context, benchmark results and evaluation footnotes. Full weights promised by 4 November; results are company-reported.
- Lumina: Kimi K3.1 Pro is now out ↗
23 October 2026. Our launch report and supplied original model-picker capture.
- Lumina: K3.1 Pro launch benchmarks ↗
23 October 2026. Our post discussing the release chart, with follow-up on Terminal-Bench and a link to the official blog.
- Lumina: single-prompt 3D train example ↗
22 October 2026. Our hands-on example shared before the launch. A product example rather than a controlled benchmark.
