Bitcoin, ether and every other major cryptocurrency fell Friday, with many pointing the finger at a Beijing startup that offered a model that beat Anthropic’s best at writing code.
Moonshot AI released Kimi K3 on Thursday, and by Friday morning AI and semiconductor stocks were selling off across Asia, with traders dubbing it a “Kimi moment,” an echo of the DeepSeek shock that wiped about $600 billion off Nvidia’s market value in a single session.
What Moonshot has built is a 2.8 trillion parameter model with a pop-up of one million tokens, about four times the size of its previous version. It operates on a mixed-expert design, meaning it only activates a small part of itself for any given task, in this case 16 out of 896 specialists, which explains why such a large model remains cheap to run.
Moonshot claims that the architectural changes give it approximately 2.5x better scaling efficiency than its predecessor, according to its tech blog.
In Arena’s Frontend Code leaderboard, K3 scored 1,679 compared to 1,631 for Anthropic’s Claude Fable 5 and 1,618 for OpenAI’s GPT-5.6, taking the top spot and ranking first in six of the seven categories.
The previous Moonshot model was at number 18. That’s a jump of 17 places in just one release. In broader tests of general knowledge work, K3 falls behind the top Claude and OpenAI setups rather than ahead of them, so this is a win in one specific area, not across the board.
