K3 forms the basis of why the list is different than it would have been a month ago. The open-weight model outperformed all rivals – except Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 – on some metrics, and surpassed a widely watched coding benchmark, triggering a semiconductor sell-off on Friday that dragged the crypto lower.
Moonshot does not move forward alone, however. China’s Alibaba announced Sunday that its Qwen3.8 model will become open, a 2.4 trillion parameter system that the company says is behind only Fable 5 among frontier models. A preview version, Qwen3.8-Max, is already available in Alibaba’s developer tools.
Open weights allow anyone to run a model without paying its manufacturer, putting pressure on the pricing power of U.S. suppliers who charge per token.
A parameter is one of the internal dials that a model adjusts during training to better predict text. Modern models contain billions, if not trillions, and this count is the rough and imperfect shorthand the industry uses for gross size.
Bitcoin traded throughout the month as a proxy for the AI financial cycle, and its miners transformed into owners of AI data centers whose leases depend on compute demand.
The next reading will be this week, when Alphabet, Tesla and Intel release results that will show whether capital spending on AI continues to rise and whether miners betting on it are staying the course.
