Member-only story
The AI race has split into two races.
Closed labs are still trying to own the smartest model you can call. Open-weight labs are trying to put a capable model inside every cloud, every inference stack, every agent framework, and eventually every company that does not want its AI product to depend on one API.
That second race is moving faster than people realise.
I do not mean that open models have cleanly beaten the closed frontier. The independent evidence does not support that. Epoch AI estimated that the strongest open-weight models were still about four months behind the best closed models through late May 2026. Stanford’s 2026 AI Index reported a 3.3 percentage-point open versus closed gap in March.
But capability is only one way to win.
A model can lose the top leaderboard row and still win distribution. If thousands of builders can download it, quantize it, put it behind their own API, fine-tune it, host it in a regulated country, or make it work with the tool stack they already use, the original lab has put itself inside a much larger part of the market.
That is why the words matter. Closed source, open weights, and open source AI are three different strategies. Calling all of them “open…