Open-source models look like they've caught up to the closed frontier. On the public benchmarks, the gap is down to a rounding error. But the benchmarks are measuring the wrong thing. The frontier labs have quietly moved the fight to ground open source can't reach: proprietary data they pay experts billions a year to create, feedback loops from real enterprise use, and models they increasingly protect with lawyers and export controls. The gap isn't closing. It's moving somewhere you can't see it. The uncomfortable corollary for enterprises: every day you run your proprietary way of doing things through a model you don't own, you're teaching someone else's system to do your job. The opportunity, and why we’re investing in the layer that lets you keep it.