Chinese AI model Kimi K2 undercuts rivals with low prices

In a fresh twist to the growing AI rivalry, Alibaba-backed startup Moonshot has unveiled its latest large language model, Kimi K2, designed to take on the likes of OpenAI’s ChatGPT and Anthropic’s Claude Opus 4.

Released late Friday, Kimi K2 not only claims to match or exceed US-based models in performance but is significantly cheaper and fully open-source, according to reports. The launch couldn’t have been more timely. Just hours after Kimi K2’s release, OpenAI’s CEO, Sam Altman, announced a fresh delay to their first open-source model, again citing safety concerns, a move that further spotlights the contrasting strategies between Chinese and American AI firms.

Moonshot’s Kimi K2 is built for code

Kimi K2’s standout feature is its coding ability. Many companies are increasingly looking to generative AI as a cost-effective way to assist or even replace some developer tasks. Moonshot is clearly aiming for that slice of the market.

In its announcement posted on X and GitHub, Moonshot claimed that Kimi K2 outperformed Anthropic’s Claude Opus 4 in two key benchmarks and also surpassed OpenAI’s GPT-4.1 in several areas. AI analyst Wei Sun from Counterpoint noted that the model is “globally competitive”, and emphasized that being open-source gives it even more appeal.

“On top of that, it has lower token costs, making it attractive for large-scale or budget-sensitive deployments.”

Wei Sun.

It is not just performance where Kimi wins, its cost too. The model is available completely for free through Kimi’s app and web browser. And for commercial use, it is hard to beat; 15 cents per million input tokens and $2.50 per million output tokens. By contrast, Claude Opus 4 charges $15 and $75, respectively, while GPT-4.1 sits at $2 for input and $8 for output tokens.

Moonshot says Kimi K2 is open to all developers, with very few restrictions. But there is only one major requirement. Commercial services that use Kimi and have over 100 million monthly users or make $20 million per month must include a “Kimi K2” credit on the interface.

The model has been met with generally positive feedback, though some users on Chinese and English social media have flagged hallucination issues, a known problem with all major generative AI tools. Pietro Schirano, founder of AI startup MagicPath, wrote on X that K2 is “the first model I feel comfortable using in production since Claude 3.5 Sonnet.”

Kimi K2 is so good at tool calling and agentic loops, can call multiple tools in parallel and reliably, and knows "when to stop", which is another important property.

It's the first model I feel comfortable using in production since Claude 3.5 Sonnet. pic.twitter.com/TcEkPlBMuk

— Pietro Schirano (@skirano) July 13, 2025

Moonshot is rising in crowded Chinese AI scene

Moonshot has been building its presence as a ChatGPT alternative in China since early last year. While rivals like ByteDance and Tencent have also entered the generative AI race, Moonshot’s focus on quality and accessibility has earned it loyal users. Even Baidu is incorporating more AI tools into its core search engine.

Moonshot’s earlier AI releases were already open-source, a rarity in China’s tech landscape. And now, with Kimi K2, the company has made a more global push, especially as US companies hesitate on fully open models.

Other Chinese firms are also making moves. DeepSeek, which made waves earlier this year, has yet to release major updates. Meanwhile, Manus AI, another promising upstart, recently shifted its base to Singapore, possibly looking for a more neutral base amid rising geopolitical tension.

Kimi K2 isn’t Moonshot’s only recent success. The company launched another research-focused model last month, which reportedly performed impressively on the “Humanity’s Last Exam” benchmark, matching Google’s Gemini Deep Research score of 26.9 and surpassing OpenAI’s performance on the same test.

It even caught attention during Elon Musk’s release of Grok 4, which scored lower unless enhanced with extra tools and web access.

Winston Ma, a professor at NYU School of Law and author of The Digital War, described Moonshot’s new researcher model as a “paradigm shift.”

He pointed to its ability to reason independently and complete complex tasks, a level of autonomous thinking not often seen in earlier large language models.

“Instead of merely generating fluent responses, it demonstrates autonomous reasoning at an expert level — the kind of complex cognitive work previously missing from LLMs,” said Ma.

As US tech giants struggle to balance innovation and safety, China’s new AI players, especially Moonshot, are stepping into the spotlight with cheaper, open, and increasingly powerful tools.

Cryptopolitan Academy: Coming Soon - A New Way to Earn Passive Income with DeFi in 2025. Learn More