Follow ZDNET: Add us as a favorite source On Google.
ZDNET Highlights
- Moonshot released its new Kimi K2 thinking model on Thursday.
- It claims to outperform GPT-5 and Sonnet 4.5 on some benchmarks.
- Open-source AI poses a challenge to proprietary US models.
The global AI arms race remains in constant flux, this time thanks to the arrival of a new model from rising Chinese AI lab Moonshot.
Also: Why open source can’t survive the rise of generic AI
The Beijing-based company on Thursday Kimi K2 Thinking releasedA reasoning model it says outperforms OpenAI’s GPT-5 and Anthropic’s Cloud Sonnet 4.5 on key benchmarks, including Humanities’ Last Exam, BrowseComp (which tests the ability of AI agents to extract hard-to-find online information via a web browser), and SEAL-0 (which assesses reasoning abilities). Kimi K2 Thinking also showed coding capabilities that were comparable to GPT-5 and Sonnet 4.5, but not particularly impressive.
“By actively reasoning using a variety of tools, K2 Thinking is able to adopt hundreds of steps of planning, reasoning, executing, and thinking to tackle some of the most challenging academic and analytical problems,” Moonshot writes on its website.
What does Kimi K2 offer
Kimi K2 Thinking is a mixture-of-experts (MoE) model that blends long-horizon planning, adaptive reasoning, and the use of online tools (such as browsers) to “continuously generate and refine hypotheses, verify evidence, reason, and construct coherent answers,” the company wrote. “This interconnected logic allows it to decompose vague, open-ended problems into clear, actionable sub-tasks.” It was trained with approximately 1 trillion parameters and can be accessed hugging face,
Also: The best free AI for coding in 2025 – only 3 available now
Crucially, KM K2 thinking – which is based on kmi k2 model Released in July – is open source, meaning developers can access the underlying code and builds for free. Read this again: A model that (according to Moonshot) has more advanced agentic capabilities than OpenAI and Anthropic’s Frontier models is free. Moonshot also said the training cost was less than $5 million — $4.6 million to be exact, according to cnbc – A vanishing sum compared to the billions spent by the most prominent AI labs in the US
If externally verified, its implications could be huge – or spark like the DeepSeek-induced panic of January 2025.
ideas for businesses
First and foremost, there is the business side of things. Since the advent of ChatGPT three years ago, there has been increasing pressure on business owners to incorporate new AI tools, particularly agents, which tech developers have marketed as productivity enhancers and virtual assistants. This often means paying for business-level offerings, like OpenAI’s ChatGPT for enterprises.
(Disclosure: ZDNET’s parent company Ziff Davis filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in the training and operation of its AI systems.)
Until now, the usual sales pitch in Silicon Valley has been that paying for a proprietary AI tool from a major developer is worthwhile, because – to paraphrase what has become a popular marketing trope – even if AI doesn’t put you out of business, another company using AI almost certainly will (never mind the fact that most businesses using AI have seen no measurable ROI).
Also: As OpenAI hits 1 million business customers, could the AI ROI trend finally turn?
Like DeepSeek’s R1, the arrival of Moonshot’s new model calls the entire logic of that sales pitch into question. Suddenly, businesses have a free AI model that is believed to be better at performing critical agentic tasks than the best proprietary models available.
Of course, it’s highly unlikely that a large number of businesses will throw out the AI baby with the bathwater and immediately cancel their OpenAI or Anthropic Enterprise subscriptions just because the latest hotshot Chinese firm claims to be creating a more advanced model. But it will definitely get some attention and make people think again: maybe the proprietary, subscription-based model of AI they’ve been sold isn’t the only way of the future.
In fact, it’s already happening: Some US companies now prefer Airbnb like AI tools from Chinese companies, compared to their American counterparts, cite both their superior performance in certain key tasks as well as their lower cost. Of course, some experts have expressed concern that open-source models, especially those of foreign origin, pose additional security risks; Several US agencies and other countries swiftly banned DeepSeek.
AI Faceoff: US vs. China
If the arrival of R1 in January was that country’s “Sputnik moment,” then the arrival of Moonshot’s KM K2 model on Thursday is the Chinese AI industry’s moon landing (in reality).
Also: AI agents are only as good as the data they’re given, and that’s a big issue for businesses
American policymakers and tech pundits have generally framed that race as an ideological one, with “American AI” on the one hand embodying the ideals of Western liberal democracy and “Chinese AI” on the other, representing centralized control over information flows and censorship.
While it is true that some AI models created by Chinese labs display bias and censor information that appear to align with the official policies of the Chinese Communist Party, it is important to keep in mind that all AI systems – no matter where their parent companies are located – come with some form of bias; The technology you use will to some extent reflect the worldview of the people creating it and the biases inherent in the data used to train it.
In any case, if the performance of the new Kimi model lives up to the impressive metrics on Moonshot’s website, ideological concerns may take a backseat to financial concerns. No investor can ignore that modest $4.6 million price tag.
Also: I’m Testing the Top AI Browsers – Here’s Which Browsers Really Impressed Me
Here in the US, while businesses and individual consumers have been sold the idea that it’s worth paying for a top-tier ownership model, investors have been sold the story that to build those devices, companies need to spend huge sums, up to tens of billions of dollars, even though many of those companies aren’t profitable yet.
So far, it’s working. Leading US AI labs like OpenAI and Anthropic are now valued in the hundreds of billions, and their spending on the infrastructure and compute needed to build increasingly advanced models is growing by the day. But fears are growing about the possibility of an AI bubble: the possibility that a large part of our global economy will be tied to a commodity that may ultimately not be able to generate profits, and which could collapse the entire house of cards, as the widespread use of securitized derivatives did in the housing market in 2008.
Also: Gartner drops its 2026 tech trends — and it’s not all AI: here’s the list
Only time will tell if we are truly living inside an AI bubble. But one thing is certain: The sudden arrival of a free tool that outperforms OpenAI and Anthropic’s leading models will have many tech investors watering their eyes – and wondering whether they should be backing a different horse.
