Moonshot AI launches Kimi K3.1 Pro, a 3 trillion parameter model with 1M context

Moonshot AI has launched Kimi K3.1 Pro, the first model in its K3.1 series and its most powerful yet, with 3 trillion parameters and 1 million token context. A highspeed version runs 2 to 3 times faster, and full weights will be out before November 4. On the Terminal-Bench 4.0 coding test, K3.1 Pro scored 65.7%, beating GPT-6 Astra and coming very close to Claude Opus 5.5. In the internal long-horizon coding test, it scored 61.2%, up from K3's 55.95%. The model is strong at 3D creation and spatial reasoning. It can turn text or sketches into usable 3D models and run scripts in tools like Blender and CAD. A new 3D Make feature is live, and the model hit a best-ever 64.4 on the OfficeQA Pro v2 finance benchmark. Stress tests show its single-fact error rate dropped from 23.4% in K3 to 8.3%. API pricing starts at 1.3 yuan per million input tokens, with output at 68 yuan.