Chinese AI company Moonshot AI has released Kimi K3.1 Pro, an upgraded version of its Kimi K3 model, in its consumer apps and for developers. In its launch post, Moonshot calls it its most capable “3T-class” open-weight model, meaning it sits in the roughly three-trillion-parameter size range (parameters are the internal settings a model learns during training). The company says the full model weights will be published by November 4, 2026.
What is new
K3.1 Pro is not a brand-new model built from scratch. Moonshot describes it as Kimi K3 with much more “post-training”, the stage after initial training where a model is tuned to follow instructions and get better at specific jobs. It keeps a 1-million-token context window, which roughly means it can take in a very long document or a large codebase in one go.
There is also a faster version, K3.1 Pro Highspeed, which Moonshot says runs two to three times quicker.
How it compares, according to Moonshot
All of the following figures come from Moonshot itself and have not been independently checked:
- On Terminal-Bench 4.0, a test of how well AI handles command-line computer tasks, K3.1 Pro scored 65.7%. Moonshot lists OpenAI’s GPT-6 Astra at 58.2% and Anthropic’s Claude Opus 5.5 at 66.4%.
- On an internal hallucination stress test, the rate of made-up claims dropped from 23.4% with K3 to 8.3%. Moonshot stresses this is a deliberately hard test, not a typical everyday error rate.
- It posted gains over K3 on several finance and office-work tests.
Moonshot also says it added measures to reduce misuse for hacking, and admits the model still trails the best closed models from US labs on some cybersecurity tests.
Where you can use it
Moonshot says K3.1 Pro is available on the Kimi website and in the Kimi apps for iPhone, Android and HarmonyOS, as well as in its Kimi Work desktop app for Windows and Apple silicon Macs and its Kimi Code tool. Developers can reach it through the Kimi API. For business accounts, it is switched off by default until an administrator turns it on.
The launch post does not say whether the model is free in the app or limited to paid plans, and it does not list prices.
Why open weights matter
“Open weights” means the trained model files are released for anyone to download, run on their own hardware and modify. That is different from models like ChatGPT or Claude, which you can only use through the company’s own service. Chinese labs, including Moonshot and DeepSeek, have leaned heavily on this approach, which puts pressure on US companies that keep their best models closed. A model of this size still needs serious data-centre hardware to run, so this is mainly relevant to businesses and researchers.
What it means for you
If you are curious, you can try K3.1 Pro in the Kimi app or website now; check the app for any plan limits, since Moonshot has not spelled them out. As with any AI service, think about what personal or work data you share. Moonshot says it offers zero data retention for API customers, but that promise is aimed at developers, not regular app users. For most people, the bigger story is that free-to-download models keep closing the gap with the top paid ones.
Sources
- Kimi K3.1 Pro — Moonshot AI