Chinese startup Moonshot AI has introduced its latest model, Kimi K3, which it claims narrows the performance gap with the top U.S. AI systems and even outperforms offerings from OpenAI and Anthropic in certain benchmarks.
Though K3 does not yet match the overall capabilities of Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol, the company announced on Friday that it consistently excelled against other models in the evaluation process.
K3 outperformed Claude Opus 4.8 and GPT 5.5—models that are just below the leading technologies from Anthropic and OpenAI—on multiple performance metrics, including coding tasks and general intelligence tests, according to Moonshot's findings.
With 2.8 trillion parameters, K3 stands as the largest AI model developed in China to date, reflective of its extensive neural network architecture.
Bank of America analysts, led by Alex Liu, noted that "despite ongoing challenges with hardware and computing capacity in China, K3 illustrates how advancements in pre-training scaling and architectural design can yield significant improvements for top-tier Chinese AI models."
This launch arrives amid a growing competition for AI dominance between the U.S. and China.
Chinese AI models are increasingly attracting interest from companies in the West, as they improve performance while remaining more cost-effective compared to the most advanced alternatives available from U.S. laboratories. Concurrently, American lawmakers are exploring ways to limit the rising incorporation of Chinese AI technologies by domestic firms.
