Core Viewpoint - Kimi Intelligent Assistant has launched its latest model Kimi K2.5, which integrates with AI programming tools from major companies like Alibaba, Tencent, and ByteDance, showcasing significant improvements in various tasks including coding, image processing, video handling, and general intelligence tasks [1] Group 1: Model Performance - Kimi K2.5 has achieved a weekly API call volume of 1.3 trillion tokens on the global API aggregation platform OpenRouter, surpassing models like Gemini3 and Claude Sonnet4.5 [1] - The model supports both conversational and agent task modes and is now available for developers [1] - Kimi K2.5 has received the best scores in multiple agent evaluations, including HLE (Human Last Exam), BrowseComp, and DeepSearchQA, among global open-source models [1] Group 2: Technical Features - The model is designed with a native multimodal architecture, allowing it to process both visual and textual inputs simultaneously [1] - Kimi K2.5 integrates capabilities for visual understanding, reasoning, programming, and agent tasks into a single model [1] - The founder and CEO of Kimi, Yang Zhilin, stated that the company has restructured the reinforcement learning infrastructure and optimized training algorithms to ensure maximum efficiency and performance [1]
Kimi K2.5接入阿里、腾讯、字节跳动AI编程工具
Feng Huang Wang·2026-02-10 08:43