An AI panel at MWC Shanghai Tech Show on June 19, 2025.
Bloomberg | Bloomberg | Getty images
Beijing – The latest model of Chinese generative artificial intelligence to face the Openai Chatppt offers coding capacities – at a lower price.
Alibaba back to back startup Released late Friday evening Its KIMI K2 model: a large open source model at low cost – the two factors that supported the disturbance of the Deepseek industry based in China in January. Open Source technology provides free access to the source code, an approach that few American technology giants have adopted Meta and Google to some extent.
By coincidence, the CEO of Openai Sam Altman Announced early Saturday that there would be an indefinite delay in its first open source model again due to security problems. OPENAI did not immediately respond to a request for CNBC comments on Kimi K2.
One of Kimi K2’s forces is the writing of the IT code for applications, an area in which companies see the potential to reduce or replace staff with a generative AI. Openai rival anthropic coding with its Claude Opus 4 model released at the end of May.
In his announcement of publication on social media platforms X and GitHub, Moshot said that Kimi K2 exceeded Claude Opus 4 on two benchmarks, and had better overall performance than the GPT-4.1 model focused on the coding of Openai, based on several metrics in the industry.
“Without a doubt [Kimi K2 is] A global model, and it is open, “an email said on Monday.
Cheaper option
“In addition to that, it has lower token costs, which makes it attractive for large-scale or budgetary deployments,” she said.
The new K2 model is available via the Kimi application and browser interface for free unlike Chatgpt or Claude, which charge monthly subscriptions for their latest AI models.
Kimi also only invoices 15 cents for each 1 million entry tokens and $ 2.50 per million exit tokens, according to his website. Tokens are a means of Measure the data for the processing of the IA model.
On the other hand, Claude Opus 4 invoice 100 times more For contributions – $ 15 per million tokens – and 30 times more for production – $ 75 per million tokens. Meanwhile, for each million tokens, Costs GPT-4.1 $ 2 for entries and $ 8 for the release.
Moshot ai said on github Whether the developers can use K2 as they wish, with the sole requirement that they display “Kimi K2” on the user interface if the product or sales service has more than 100 million monthly active users, or makes the equivalent of $ 20 million in monthly income.
Hot ia market
K2 initial criticisms on English and Chinese social media have been largely positive, although there is Some hallucinations reportsA widespread problem in the generative AI, in which the models make up information.
However, K2 is “the first model that I feel comfortable using production since Claude 3.5 Sonnet”, Pietro Schirano, Startup Magicpath founder This offers AI tools for design, said in an article on X.
Moonshot has opened some of its previous AI models. The company’s chatbot increased in popularity at the beginning of last year as an alternative from China to Chatgpt, which is not officially available in the country. But similar chatbots from Bytedance and Tencent have since closed the market, while the Baidu technology giant has reorganized its basic search engine with AI tools.
The latest version of Kimi’s AI comes when investors have Chinese alternatives to American technology in the AI global competition.
However, despite the excitement about Deepseek, the private company has not yet announced a major upgrade of its R1 and V3 model. Meanwhile, Manus Ai, a Chinese startup that emerged earlier this year like any other timeless type. moved his seat in Singapore.
In the United States, Openai has not yet revealed GPT-5.
The work on GPT-5 can take engineering resources, preventing OpenAI from progressing on its open source model, said Sun de Counterpoint, adding that it is difficult to publish a powerful open source model without undergoing the competitive advantage of a owner model.
Grok 4 Competitor
Kimi K2 is not the only recent version of the company. Moonshot launched a Kimi research model last month and said it corresponded to the GOOGLE 26.9 score in depth and Beat the version of Openai on a reference called “Last exam of humanity”.
The Kimi research model was even mentioned last week during Elon Musk XAI Liberation of Grok 4 – which marked 25.4 alone on the reference of “last examination of humanity”, but has reached a score of 44.4 when authorized to use a variety of AI and web search tools.
“Kimi-researcher represents a paradigm shift in the agency’s AI,” said Winston Ma, auxiliary professor at the Nyu School of Law. It referred to the AI capacity to simultaneously make several decisions to accomplish a complex task.
“Instead of simply generating current responses, he demonstrates autonomous reasoning at an expert level – the type of complex cognitive work missing before in the LLM”, said Ma. He is also the author of “The Digital War: How China’s Tech Power shapes the future of AI, Blockchain and Cyberespace”.
– Victoria Yeo de CNBC contributed to this report.
