MoNshot has presented today As k2an agentic artificial intelligence that, he says, surpasses the latest OpenAi and Deepseek models.
As k2 It is an open source language model that the Chinese company calls «mixture of experts», with 32,000 million activated parameters and a total of 1 billion parameters.
Little by little, China is also reaching the United States in AI, without the need to consume so many energy and hardware resources. Deepseek caused a real earthquake a few months ago, and now MoNshot wants to do the same with Kimi K2.
Kimi K2, China that aims to be better than chatgpt
«Kimi K2 reaches premium performance in knowledge of avant -garde, mathematics and coding between non -thinking models. But goes further: meticulously optimized for agentive tasks, Kimi K2 not only responds, but acts,» says the press release.
Kimi K2 Chatea, answers doubts, analyzes, generates AI agentsand it is also very good writing code. It is designed for developers and researchers, and for standard users.
In the benchmarks that MoNshot has published, overcomes GPT-4.1 and Deepseek v3 in aspects such as code generation and AI agentsuse of tools, and mathematics and stem. With Claude 4 it is a little more equal. You can see it in these graphics:
Moonshot
Kimi K2 is offered in two versions. Who-k2-base It is the standard model, especially thought for researchers and developers who want total control for fine adjustment and personalized solutions.
Like-K2-INSTRUCT It is the ideal post-designed model for chat experiences and agents of general use. It is a reflex degree model without thinking much.
You can try free K2 free and without registrationdirectly from its website. Although he is in English, you can ask in Spanish, he understands and will answer you in this language.
The AI race, for the moment, is a hand in hand between China and the United States. And it is a career where to be second is to fail. As k2 He presents his candidacy to overcome Chatgpt.
Know How we work in NoticiasVE.
Tags: Artificial Intelligence, Industry, China