Moonshot AI has launched its new flagship large language model, Kimi K3.1 Pro. This model is an upgraded, post-trained version of the Kimi K3 algorithm, which was originally released in July and features 3 trillion parameters across 896 neural networks.
The Kimi K3.1 Pro model demonstrates high performance in 3D modeling, achieving a score of 92.7% on the BenchCAD benchmark. Beyond static models, the AI is capable of generating animated simulations and complex 3D video games.
In coding and analytical tasks, the model scored 65.7% on the Terminal-Bench 4.0 benchmark and outperformed competing models on the OfficeQA Pro v2 test, which evaluates financial analysis capabilities.
Moonshot AI intentionally excluded cybersecurity data from the model's training set to mitigate safety risks, resulting in lower performance on cybersecurity benchmarks compared to rival models.
The model is currently available through the company's paid cloud services. Moonshot AI has scheduled the release of the model's open-source weights for November 4.
