AI sign for the MWC Shanghai Institute of Technology Show on June 19, 2025.
Bloomberg | Bloomberg | Getty Images
Beijing – The latest Chinese Generated Artificial Intelligence model undertakes Openai’s ChatGPT offers coding capabilities at a lower price.
AlibabaBacked Startup Moonshot released the Kimi K2 model late on Friday night: a leading low-cost open source language model – two factors that helped disrupt the Chinese-based Deepseek industry in January. Open source technology offers free source code access, an approach rarely taken by US tech giants that take anywhere other than meta and Google.
Coincidentally, Openai CEO Sam Altman announced earlier on Saturday that there will be another indefinite delay in the first open source model due to safety concerns. Openai did not immediately respond to a CNBC request for comments about Kimi K2.
One of the strengths of Kimi K2 is that it writes computer code for the application. This is an area where companies consider the possibility of reducing or replacing staff with AI generated. Openai’s US rival humanity focused on coding on the Claude Opus 4 model, which was released in late May.
In the announcement of its release on social media platforms X and Github, Moonshot claimed that Kimi K2 outperforms Claude Opus 4 on two benchmarks, and that it performs better overall than Openai’s coding-centric GPT-4.1 model based on several industry metrics.
“no doubt [Kimi K2 is] Wei Sun, principal analyst for artificial intelligence at Counterpoint, said in an email Monday.
Cheaper options
“Additionally, it’s attractive for low token costs and large or budget-sensitive deployments,” she said.
The new K2 model is available for free, unlike ChatGPT and Claude, via Kimi’s app and browser interface.
You’re also charged only $2.50 per million output tokens, according to its website. Tokens are methods of measuring data from AI model processing.
In contrast, the Claude Opus 4 charges 100 times the input ($15 per million) and over 30 times the output – $75 per million. Meanwhile, for every million tokens, GPT-4.1 charges $2 for input and $8 for output.
Moonshot AI said on GitHub that developers can use K2, but there is the only requirement to display “Kimi K2” in the user interface if your commercial product or service has more than 100 million active users per month, or if your monthly revenue is worth $20 million.
Hot AI Market
Although K2’s initial reviews on both English and Chinese social media were largely positive, there are several reports of hallucination, a common problem with the generation AI that the model makes up the information.
Still, the K2 is “the first model I’ll use in production since the Claude 3.5 Sonnet,” Pietro Schirano, founder of Startup MagicPath, an AI tool provider, said in a post on X.
Moonshot has opened some of its previous AI models. The company’s chatbot surged in popularity early last year as an alternative to China’s ChatGPT, but it is not officially available within the country. However, similar chatbots from Bytedance and Tencent have since been busy the market, while Tech Giant Baidu has revamped its core search engine with AI tools.
Kimi’s latest AI release aims to be a Chinese alternative to US Tech in the global AI competition.
Still, despite the excitement about Deepseek, the individual-owned company has yet to announce any major upgrades to the R1 and V3 models. Meanwhile, Manus AI, a Chinese startup that appeared as another DeepSeek startup earlier this year, moved its headquarters to Singapore.
In the US, Openai has yet to reveal the GPT-5.
The work on GPT-5 covers engineering resources and that Openai may be hampering the progression of open source models, CounterPoint’s Sun added that it would be difficult to release a strong open source model without compromising the competitive advantages of its own model.
Grok 4 Rival
Kimi K2 is not the company’s only recent release. Moonshot launched the Kimi Research model last month, matching Google’s Gemini Deep Research score of 26.9, claiming it defeated Openai’s version with a benchmark called “Humanity’s Last Exam.”
Kimi’s research model was mentioned last week during the Xai release of Elon Musk’s Grok 4. This scored 25.4 on the “Humanity’s Last Test” benchmark, but with various AI tools and web searches it scored a 44.4 score.
“Kim Researcher represents a paradigm shift in agent AI,” says Winston MA, an adjunct professor at NYU Law School. He mentioned the ability of AI to make several decisions on its own to complete complex tasks.
“It not only produces fluent responses, it also shows autonomous reasoning at the expert level. This is the kind of complex cognitive task that has not been previously lacking from LLMS,” Ma said. He is also the author of “Digital Wars: How China’s technological capabilities shape the future of AI, blockchain and cyberspace.”
– CNBC’s Victoria Yo contributed to this report.
