Alibaba-backed Moonshot's new Kimi AI model beats ChatGPT, Claude in coding — and it costs less

In a bold move in the world of generative artificial intelligence, Moonshot, a startup backed by Alibaba, has introduced its latest AI model, Kimi K2, which promises enhanced coding capabilities at a significantly lower price than its competitors. Launched late Friday, Kimi K2 emerges as a low-cost, open-source large language model, a strategy reminiscent of the disruptive approach taken by DeepSeek earlier this year. Unlike many U.S. tech giants, which have been cautious about open-source technology, Moonshot's strategy allows developers free access to source code. This development comes at a time when OpenAI CEO Sam Altman announced another delay for the company's much-anticipated open-source model, citing safety concerns. Kimi K2 is designed with a strong focus on writing computer code, an area where businesses are keen to leverage generative AI to reduce staffing needs. In a competitive landscape, OpenAI's rival, Anthropic, has also emphasized coding with its Claude Opus 4 model, released in May. In its announcement on social media platforms X and GitHub, Moonshot claimed that Kimi K2 outperformed Claude Opus 4 in two key benchmarks and demonstrated superior overall performance compared to OpenAI's coding-focused GPT-4.1 model. Wei Sun, a principal analyst at Counterpoint, noted in an email, "Kimi K2 is undoubtedly a globally competitive model that is open-sourced. Its lower token costs make it particularly appealing for large-scale or budget-conscious deployments." Accessible through Kimi's app and web interface, Kimi K2 is free for users, unlike ChatGPT and Claude, which require monthly subscriptions for access. Kimi charges just 15 cents for every 1 million input tokens and $2.50 for each million output tokens—a stark contrast to Claude Opus 4's pricing of $15 and $75 per million tokens, respectively. Developers are encouraged to utilize Kimi K2 freely, provided they credit "Kimi K2" on their interfaces if their products exceed 100 million monthly active users or generate over $20 million in monthly revenue. Initial feedback from users on social media has generally been positive, although some have reported instances of hallucinations, a common issue in generative AI where models fabricate information. Pietro Schirano, founder of the startup MagicPath, expressed enthusiasm for Kimi K2, stating it is the first model he feels comfortable using in production since Claude 3.5 Sonnet. Moonshot has previously open-sourced other AI models, gaining traction early last year as China's alternative to ChatGPT, which is not officially available in the country. However, competition has intensified with the emergence of similar chatbots from companies like ByteDance and Tencent, alongside Baidu's revamped AI-integrated search engine. As investors look towards Chinese alternatives in the global AI race, Moonshot's new release comes amid significant developments in the industry. Despite the buzz surrounding DeepSeek, the company has yet to announce significant updates to its existing models. Meanwhile, Manus AI, another emerging Chinese startup, has moved its operations to Singapore. In contrast, OpenAI has yet to unveil GPT-5, with ongoing work on that model potentially hindering progress on its open-source initiatives. Kimi K2 isn't Moonshot's only recent advancement. The company launched a Kimi research model last month, claiming it matched Google's Gemini Deep Research's score and outperformed OpenAI's version on a benchmark known as "Humanity's Last Exam." This research model even gained attention during Elon Musk's xAI launch of Grok 4, which performed well under specific conditions. Winston Ma, an adjunct professor at NYU School of Law, remarked that Kimi-Researcher signifies a paradigm shift in agentic AI, highlighting its ability to undertake complex tasks through autonomous reasoning, a capability that has traditionally been lacking in large language models. Ma, who authored "The Digital War: How China's Tech Power Shapes the Future of AI, Blockchain and Cyberspace," emphasized the significance of this advancement in AI technology.