Kimi 3.1 Pro Narrows Closed-AI Gap with 3-Trillion Parameters and High-Speed API
Tencent WorkBuddy has integrated Kimi's new K3.1 Pro model along with a high-speed version offering 2-3x output speeds. K3.1 Pro is a 3-trillion parameter model supporting a 1-million context window and exceptional performance in coding, 3D creation, and financial workflows. This release significantly narrows the performance gap between open-weight models and closed-source frontier AI systems, achieving state-of-the-art results on several prominent benchmarks. The introduction of a high-speed API also improves practical deployment efficiency for enterprise and individual users. K3.1 Pro scored 65.7% on Terminal-Bench 4.0, outperforming GPT-6 Astra, and achieved top scores on OfficeQA Pro v2 and SpreadsheetBench v2. The development involved heavily increased reinforcement learning compute, and it supports robust programmatic CAD tasks via BenchCAD.
## BACKGROUND
Reinforcement learning (RL) is a machine learning training method where models learn by trial and error, rewarded for desirable behaviors to enhance complex reasoning. Benchmarks like Terminal-Bench and BenchCAD evaluate specialized agent capabilities such as terminal command execution and multi-view 3D model reconstruction from code.