Kimi K3 is not cheap (www.alexinch.com)

The recent unveiling of Kimi K3, a new open-weights large language model (LLM) from China's Moonshot AI, has created a stir within the AI/ML community. While some analysts claimed K3 could rival U.S. models at a lower cost, a deeper examination reveals that this assumption is misleading. In fact, K3 is relatively expensive, costing slightly less per task than OpenAI's GPT-4 model but significantly more than other Chinese LLMs, such as GLM-5.2, which it costs double, and is approximately 20 times pricier than DeepSeek V4. Despite its higher price point, K3 represents a substantial advancement for Chinese LLMs and the open-weights paradigm. Its ability to be fine-tuned with just a few GPUs adds to its appeal, especially in coding tasks, though it struggles with more straightforward office applications due to its verbosity. This development raises concerns for the U.S. AI market, as it indicates that Chinese labs like Moonshot AI are quickly advancing and potentially benefiting from distillation techniques, which could reshape the competitive landscape in AI technology.