Chinese startup Moonshot launches Kimi K3 model that rivals leading US AI systems
Just the facts

Chinese startup Moonshot launches Kimi K3 model that rivals leading US AI systems

Summary

Moonshot's Kimi K3 large-language model performed at the top of Arena's coding benchmark, prompting comparisons with Anthropic's Claude and OpenAI's ChatGPT, while Chinese officials highlighted the need for global AI cooperation.

Beijing-based AI startup Moonshot unveiled its Kimi K3 large-language model on Friday, a development that has drawn attention from U.S. tech firms. In Arena's benchmark for front-end coding capability, Kimi K3 ranked first, a result the platform’s co-founder and CEO Anastasios Angelopoulos described on social media as possibly "the single biggest release of the year." Angelopoulos added that additional results were expected to confirm the model’s leading position.

The launch coincided with Chinese President Xi Jinping’s opening address at the World Artificial Intelligence Conference in Shanghai, where he emphasized that AI development should involve "a symphony of global cooperation." Moonshot, which partners with Huawei, did not disclose the hardware used for Kimi K3, but the company is known to collaborate with the Chinese tech giant.

Bank of America analysts reported that Kimi K3’s usage fees are the highest among Chinese AI models yet remain roughly half the cost of OpenAI’s GPT-5.6 Sol offering. The model follows last month’s release of Zhipu’s GLM-5.2, which has been adopted by developers for its performance-to-price ratio.

U.S. AI companies, including Anthropic and OpenAI, have previously alleged that Chinese firms engage in "distillation"—training smaller models on outputs from larger ones—to replicate proprietary capabilities. Anthropic’s February statement accused several Chinese labs of illicitly extracting Claude’s capabilities, a claim Beijing has denied.

Moonshot’s co-founder and CEO Yang Zhilin, a Carnegie Mellon University Ph.D. graduate, has a background in machine-learning research. His former adviser, Russ Salakhutdinov, praised the achievement as a win for the open-source community. Open-source AI models, which allow public inspection and modification, are argued to foster innovation, though some experts warn they may also raise safety and security concerns.

Keep reading — for free

Create a free account to follow the news. No card required.