Credit: CFP
AsianFin -- Chinese AI startup Moonshot, backed by Alibaba, has released a powerful new open-source language model that it says outperforms major U.S. rivals in code generation—at a fraction of the cost.
Unveiled late Friday, the Kimi K2 model is Moonshot’s latest entry into the generative AI race. The company claims it surpasses Anthropic’s Claude Opus 4 and OpenAI’s GPT-4.1 in coding benchmarks, while undercutting them on price. In a weekend post on GitHub and social media, Moonshot said Kimi K2 excels in generating computer code, an area where AI is increasingly seen as a way to automate technical tasks or reduce headcount.
Kimi K2 is open-source and available to the public via app and browser interfaces, making it one of the most accessible high-performing models from a Chinese company to date. Unlike ChatGPT or Claude, which require monthly subscriptions to access their top-tier models, Kimi K2 can be used free of charge. Moonshot is also offering developers usage at ultra-low prices: $0.15 per million input tokens and $2.50 per million output tokens. That compares with Claude Opus 4’s rates of $15 and $75, respectively, and OpenAI’s GPT-4.1 pricing at $2 and $8.
Wei Sun, principal AI analyst at Counterpoint Research, said in an email that Kimi K2 is "unquestionably globally competitive" and noted that its cost efficiency makes it “particularly attractive for budget-sensitive or large-scale enterprise deployments.”
Moonshot is requiring commercial users with over 100 million monthly active users or $20 million in monthly revenue to display the “Kimi K2” name on their interface, but otherwise allows broad, unrestricted usage of the model.
Initial user feedback on platforms like X and GitHub has been largely positive, with early adopters praising its performance despite occasional hallucinations—an industry-wide issue where models generate incorrect or fabricated information. Pietro Schirano, founder of AI design startup MagicPath, said Kimi K2 was the first model he felt comfortable deploying since Claude 3.5 Sonnet.
Moonshot is one of several Chinese companies pushing aggressively into the generative AI space, especially as OpenAI’s ChatGPT remains unavailable in China. While early Chinese alternatives gained traction, competition has intensified with offerings from ByteDance, Tencent, and Baidu. Kimi’s rise reflects investors’ growing interest in Chinese challengers as global AI competition heats up.
Kimi K2 follows another Moonshot release last month—the Kimi Researcher model—which made headlines for its strong showing on “Humanity’s Last Exam,” a benchmark designed to test advanced reasoning. The research model matched Google Gemini Deep Research’s 26.9 score and outperformed OpenAI’s equivalent. It was even mentioned during the launch of Elon Musk’s Grok 4, which scored 25.4 independently and 44.4 when supported by tools and internet access.
“Kimi-Researcher represents a paradigm shift in agentic AI,” said Winston Ma, adjunct professor at NYU School of Law and author of The Digital War. “It’s no longer about generating fluent language—it’s about reasoning through multi-step decisions like a human expert.”
Despite the excitement, other Chinese AI contenders are struggling to maintain momentum. DeepSeek, which disrupted the market earlier this year, has yet to roll out major upgrades to its flagship R1 and V3 models. Meanwhile, new entrants like Manus AI have moved their headquarters to Singapore to tap international markets.
OpenAI, for its part, has yet to release GPT-5. According to Sun, its engineering resources may be tied up in that project—delaying progress on the open-source front. “Balancing innovation with competitive advantage is difficult when the stakes are this high,” she said.
With Kimi K2, Moonshot has signaled it’s ready to compete not just in China, but globally—and with pricing and accessibility at the core of its strategy, it may have found a path others are reluctant to take.
更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App
