Good question! First of all, open-source ≠ open weights. What Kimi is opening up are the weights, which saves you the most expensive part—the initial pre-training. But it’s not quite at the plug-and-play level yet. It’s like sharing the design blueprints for a Ferrari; there’s still a long way to go before you can actually produce a Ferrari. Secondly, this is a low-cost way to cultivate user habits. Just look at those overseas developers buying Kimi and GLM tokens through various intermediaries and APIs—you’ll see that without much advertising, they’ve managed to attract core users in the core community. It’s a cost consideration. Now compare that to how much Gemini and ChatGPT spent on advertising in the U.S. just to get developers to use their products. Third, I think anyone who’s actually downloaded the weights and tried deploying local AI knows that running it locally is really difficult. With all the effort it takes to tinker with it, you might as well just buy the official API directly. This applies to individuals, enterprises, and even other countries. Anyway, domestic AI prices are getting cheaper and cheaper, so the cost of maintenance and inference < the cost of directly buying ready-made Kimi K3/GLM 5.2 tokens. Lastly, ethically speaking, maybe Kimi did distill some of its competitors’ AI, which is inherently a bit of a moral low point. But then they turned around and open-sourced (the weights), which suddenly makes them look a lot more righteous . It’s almost got a Robin Hood vibe to it. So yeah, opening up the weights might also have some ethical considerations behind it.