Alibaba, who supports Alibaba, releases Kimi K2 Ai Surring Chatgpt, Claude

Beijing is the latest Chinese generative model of artificial intelligence to be taken in Openai chat, offers coding opportunities – at least.

Alibaba, supported by Alibaba, launch moonshine released at the end of Friday night Its Kimi K2 model: inexpensive, open source models-two factors that underpin China based on China DEEPEEK industry violation in January. Open source technology provides access to the source code for free, approach that used a few technological giantApart from Meta and Google to a certain extent.

Accidentally CEO of Openai Sam Altman announced at the beginning of Saturday that due to safety issues, an uncertain delay of its first model with open source will appear. Openai did not immediately respond to the CNBC request for a commentary to Kimi K2.

One of the strong sides of Kimi K2 In writing a computer code for applicationsan area in which businesses see the potential to reduce or replace staff with generative II. Openai American competitor anthropic focused on coding with its model Claude Opus 4 released at the end of May.

In his message on the issue of social media X and GitHub platforms, Moonshot claimed that Kimi K2 exceeded Claude Opus 4 on two landmarks and had higher total performance than a GPT-4.1 Openai coding model.

“Undoubtedly (Kimi K2 is a worldly competitive model, and it is open,” the chief analyst of artificial intelligence said on Monday.

Despite unprocessed enhancements in coding, Sun noted that Kimi K2 still develops tools for effective integration with other technological systems. This means that the developer can be more difficult to move from the officers to Kimi, which makes improvement of “less game change,” she said.

KIMI also charges only 15 cents for every 1 million input and $ 2.50 per 1 million tokens, according to the site. Tokens – this is a way Measurement data to process AI model.

Unlike this, Claude Opus 4 charges 100 times more To enter – $ 15 million – and 30 times more for production – $ 75 per million tokens. Meanwhile, for every million tokens, GPT-4.1 Charging $ 2 to enter and $ 8 for exit.

Moonshine AI said on GitHub The fact that developers can use K2 as they want, with the only requirement to show “Kimi K2” on the user interface if a commercial product or service has more than 100 million active users a month, or make an equivalent of $ 20 million monthly income.

Initial reviews of K2 both in English and Chinese social media were largely positive, though there are Some reports of hallucinationsThe common issue in the generative II, in which the models are information.

However, K2 is “the first model that I am comfortable using in production since CLUDE 3.5 Sonnet” Pietro Schirano, Founder Startup Magicpath This offers AI tools for design said in a message on X.

Moonshot has opened some of its previous AI models. Chatbot company has increased in popularity earlier last year as an alternative to chatgpt China, which is not officially available in the country. But similar chat -boots with bytes and tencent have since overflowing the market, while the technological giant Baida has updated its main search engine with AI tools.

The latest AI Kimi issue comes as investors in Chinese alternatives to us technology in the AI World Competition.

However, despite the excitement of Deepseek, the private company has not yet announced the serious upgrading of its R1 and V3 model. Meanwhile, Chinese Startup Manus AI, which appeared earlier this year as another Deepseek Type Upstart, moved the headquarters to Singapore.

The US Openai also has not yet disclosed GPT-5.

Working on GPT-5 can engage in engineering resources, preventing Openai progressing on an open source model, said Sun CounterPoint, adding that it is difficult to release an open source model without undermining your own model.

Kimi K2 is not the only recent company issue. Moonshot launched a Kimi study model last month and claimed it was in line with Google Gemini Deep Research score and 26.9 points and The Openai version won the benchmark entitled “The Last Exam of Humanity.”

Last week, Kimi’s study model even mentioned Elon Musk XAI Issue Grok 4 – which scored 25.4 independently at the “Last Exam” phase, but reached the 44.4 points when allowed to use different AI tools and searches on the Internet.

“Kimi-result is a shift of paradigms in Agentic AI,” said Winston MA, professor of the New York Law School school. He meant the possibility of II simultaneously making several decisions on his own to complete a difficult task.

“Instead of just creating free answers, it demonstrates autonomous reasoning at the expert level – a kind of complex cognitive work that was previously absent in LLM,” MA said. He is also the author of “Digital War: how China’s technological power forms the future AI, Blockchain and Cyberspace”.

– Victoria Yeo made this report in this report.