The Dark Side of the Moon officially launched the K3.1 Pro, the first model in the Kimi K3.1 series, on October 24. This model is the company's most powerful 30 trillion parameter model to date, supporting 1 million contexts. The high-speed version, Kimi K3.1 Pro Highspeed, was also launched simultaneously, offering output speeds approximately 2 to 3 times faster at the same intelligence level.
According to official sources, the K3.1 Pro achieves or approaches the performance level of the strongest closed-source models in code, agent, and knowledge benchmarks. In the Terminal-Bench 4.0 evaluation, the K3.1 Pro scored 65.7%, surpassing GPT-6 Astra's 58.2%, and is only slightly behind Claude Opus 5.5 at 66.4%. In the internal benchmark Kimi Code Bench 3.0 Long-horizon, the K3.1 Pro scored 61.2%, higher than K3's 55.95%.
K3.1 Pro Excels in 3D Creation and Spatial Reasoning
In terms of 3D creation and spatial reasoning, the K3.1 Pro achieved a score of 92.7% in the BenchCAD benchmark test, close to Claude Opus 5.5's 96.2% and GPT-6 Astra's 95.9%. The model can infer the shape, proportion, and spatial relationships of objects based on text descriptions, sketches, or reference images, and reconstruct them into usable 3D models. The 3D Make feature has been launched on kimi.com and the Kimi App, supporting online preview, editing, and 3D printing exports.
In financial workflows, the K3.1 Pro scored 64.4 in the OfficeQA Pro v2 benchmark test, achieving the best results to date. In three internal financial benchmarks—financial modeling, financial research, and business operations—the K3.1 Pro improved by 14%, 20%, and approximately 4.2 times, respectively, compared to K3. In terms of factual accuracy, the single factual statement error rate in the internal stress test Kimi Hallu Bench Hard dropped from K3's 23.4% to 8.3% for the K3.1 Pro.
Starting today, users can access the K3.1 Pro and the high-speed version through kimi.com, Kimi App, Kimi Work, Kimi Code, and Kimi API. The API pricing is set at 13 yuan per million tokens for input (cache miss), 1.3 yuan (cache hit), and 68 yuan for output; the high-speed version is priced at double that. The complete model weights will be released by November 4.