Китайцы открыли веса Kimi K3.1 Pro. Модель уже обходит GPT-6 и Claude в отдельных тестах Moonshot AI выложила Kimi K3.1 Pro на Hugging Face. Модель содержит 2,8 трлн параметров, из которых при обработке токена активируются 104 млрд. Контекстное окно - 1 млн токенов. По опубликованным бенчмаркам результаты интересные: Terminal-Bench 4.0: Kimi - 65,7, Claude Opus 5.5 - 66,4, GPT-6 Astra - 58,2. SWE Marathon: Kimi - 51,7, Claude - 46,3, GPT-6 Astra - 42,5. BrowseComp: Kimi - 92,1, GPT-6 Astra - 91,5, Claude - 90,8. Модель поддерживает изображения, программирование, исследовательские задачи и работу ИИ-агентов. Для длинного контекста используется механизм Gated DeepSeek Sparse Attention. Веса доступны в формате MXFP4, модель можно разворачивать через vLLM и SGLang. Размер файлов составляет около 1,56 ТБ, поэтому для локального запуска потребуется серьёзное оборудование. https://huggingface.co/moonshotai/Kimi-K3.1-Pro
From the original source:
Telegram — AI Machine Learning Big Data
Read full article at Telegram — AI Machine Learning Big DataThe Chinese have discovered the weights of Kimi K3.1 Pro. The model already beats GPT-6 and Claude in some tests
The Chinese have discovered the weights of Kimi K3.1 Pro. The model is already beating GPT-6 and Claude in separate tests. Moonshot AI posted Kimi K3.1 Pro on Hugging Face. The model contains 2.8 trillion parameters, of which 104 billion are activated when processing a token. Context window - 1 million tokens. By…
AI Summary & Key Takeaways
- The Chinese have discovered the weights of Kimi K3.1 Pro
- The model is already beating GPT-6 and Claude in separate tests
- Moonshot AI posted Kimi K3.1 Pro on Hugging Face
The Chinese have discovered the weights of Kimi K3.1 Pro. The model already beats GPT-6 and Claude in some tests
Related stories
Anthropic's text watermark is easy to bypass on several older Claude models
Anthropic is rolling out text watermarks to four more Claude models ahead of the EU AI Act's December deadline. But on older models like Claude Opus 4.6 and Sonnet 4.6, API users can disable the watermark by changing a single sampling parameter. Anthropic's response to the…
Texas was the new epicenter of the AI data center buildout. That might be kaput.
Texas is the fastest-growing data center market in the US and ground zero for Big Tech’s largest AI buildouts. But the blue wave that swept the Lone Star State may jeopardize that position. Democrat James Talarico won the US Senate seat against Ken Paxton. Gina Hinojosa won the…
Building on open source AI? Hugging Face’s policy chief on the costs, risks, and trade-offs
Open source models are becoming more mainstream as enterprises, governments and startups alike rely on the promise of lower costs, more control, and more flexibility. But as increasingly capable models become more freely available, how do companies capture value? And with…
OpenAI acquires Pierre Computer Company to make Git keep up with AI coding agents
OpenAI is buying Pierre Computer Company, the startup whose libraries already power the coding agent Codex. Combined with OpenAI's recently launched Git backend Pando, the deal shows the company is building out its infrastructure for developers and AI agents working side by…