For a company named “Open” AI their reluctance to just opening the weights to this model and washing their hands of it seem bizarre to me.

It’s not when you understand the history. When StabilityAI released their Stable Diffusion model as an open-source LLM and kickstarted the whole text-to-image LLM craze, there was a bit of a reckoning. At the time, Meta’s LLaMA was also out there in the open. Then Google put out an internal memo that basically said “oh shit, open-source is going to kick our ass”. Since then, they have been closing everything up, as the rest of the companies were realizing that giving away their models for free isn’t profitable.

Meanwhile, the Chinese have realized that their strategy has to be different to compete. So, almost every major model they’ve released has been open-source: DeepSeek, Qwen, GLM, Moonshot AI, Kimi, WAN Video, Hunyuan Image, Higgs Audio. Black Forest Labs in Germany, with their FLUX image model, is the only other major non-Chinese company that has adopted this strategy to stay relevant. And the models are actually good, going toe-to-toe with the American close-sourced models.

The US companies have committed to their own self-fulfilling prophecy in record time. They will spend trillions trying to make profitable models and rape the global economy in the process, while the Chinese wait patiently to stand on top of their corpses, when the AI bubble grenade explodes in their faces. All in the course of 5 years.

Linux would be so lucky to have OS market share dominance in such an accelerated timeline, rather than the 30+ years it’s actually going to take. This is a self-fail speedrun.