Moonshot AI launches Kimi K3.1 Pro, an open-weight model with weights due by 4 November
On 24 October 2026, China's Moonshot AI launched Kimi K3.1 Pro, the first model in its K3.1 series, with a 1-million-token context window. It is available now in the Kimi app, Kimi Code and the Kimi API. Moonshot calls it an open-weight model but says the full weights will be released by 4 November.
The short version
- 01Kimi K3.1 Pro is live on Kimi.com, Kimi Work, Kimi Code and the Kimi API.
- 02Moonshot says it scores 65.7% on Terminal-Bench 4.0, against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra.
- 03These are Moonshot's own figures.
- 04Moonshot says it removed cyber-related data from training to reduce misuse risk.
- 05Full model weights are due by 4 November 2026.
At a glance
| Launched | 24 October 2026 |
|---|---|
| Company | Moonshot AI |
| Model | Kimi K3.1 Pro, plus a faster Highspeed version |
| Context window | 1 million tokens |
| Available | Kimi.com, Kimi Work, Kimi Code, Kimi API |
| Weights | Due by 4 November 2026 |
| Status | Confirmed |
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is the first model in Moonshot AI's new K3.1 series. Moonshot calls it its most capable open-weight model of its size class to date, and says it is a post-trained version of Kimi K3 with much more training compute. Its context window, how much text it can handle at once, is 1 million tokens.
Moonshot also released Kimi K3.1 Pro Highspeed, which it says runs 2 to 3 times faster. Both are described in Moonshot's tech blog.
How does Kimi K3.1 Pro score?
On Terminal-Bench 4.0, a test of coding agents on terminal tasks, Moonshot says Kimi K3.1 Pro scores 65.7%. It lists Claude Opus 5.5 at 66.4% and GPT-6 Astra at 58.2%. On BenchCAD, a 3D modelling test, it says Kimi scores 92.7% with tools, against 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra.
These are Moonshot's own figures, published by the company that made the model.
Does it make fewer mistakes?
Moonshot says that on its own stress test for factual accuracy, the share of false claims fell from 23.4% for Kimi K3 to 8.3% for K3.1 Pro. It built the test itself, and the figure compares K3.1 Pro with its predecessor, not with other companies' models.
What did Moonshot do about cyber risk?
Moonshot says it found cyber-related content in its training data and removed it, at two stages of training, to reduce misuse risk. On its own hacking test, it says a version without the removal solved 90.5% of tasks, and the final model solved 81.0%.
It says the final model's offensive cyber ability is below what its pipeline could have produced, and behind that of the top proprietary models.
Who can use it, and is it open?
It is available now on Kimi.com, the Kimi apps, Kimi Work, Kimi Code and the Kimi API. Moonshot says Kimi Enterprise administrators can turn it on, but it is off by default at launch.
Moonshot calls it open-weight, but says the full weights will be released by 4 November 2026. Until then they are not downloadable.
Why does it matter?
If Moonshot's numbers hold up under outside tests, an open-weight model from China would be close to the best US models on coding. Open weights also let others run and study the model themselves.
What's next?
The weights are due by 4 November 2026. Independent tests of the model will show how its scores compare.
Questions people ask
Who makes Kimi K3.1 Pro?
Moonshot AI, a Chinese AI company.
Is Kimi K3.1 Pro open-source?
Moonshot calls it open-weight. The weights are due by 4 November 2026 and are not out yet.
Is it better than Claude Opus 5.5?
On Terminal-Bench 4.0, Moonshot reports 65.7% for Kimi against 66.4% for Claude Opus 5.5. These are Moonshot's own figures.
Where can I use it?
On Kimi.com, the Kimi apps, Kimi Work, Kimi Code and the Kimi API.
What is a context window?
It is how much text a model can handle at once. Kimi K3.1 Pro's is 1 million tokens.
Sources
Why "Confirmed"? We label a story "Confirmed" when the company or government announced it itself. Spotted an error? Email hello@openloopnews.com and we will correct it.
