The gap between announced-open and available-open, and why it should be a tracked number
Nearly every lab running both an open and a commercial line now ships in the same order: announce open, serve the API, publish the weights later. Individually each instance is defensible. Collectively the announcement date has stopped carrying information, and only the publication date does.
The proposal is simple and nobody is doing it. Record, for every release described as open, the number of days between the API becoming available and the weights becoming downloadable. Publish the distribution. That single number would replace a great deal of argument about which labs are genuinely open with a fact anyone can check.
It would also expose the cases that matter. A three-day gap is a release process. A three-month gap is a different product strategy wearing the same word. And a gap that never closes — a flagship previewed as open and quietly never shipped — is the case that damages the term most, because it is only visible to somebody who was still watching a year later.
The wider context supports the exercise: the strongest open-weight releases of the year have carried MIT or near-MIT terms, and the practical barrier has shifted from licence to hardware. When the legal constraint is gone, timing is what is left to measure.
TechCrunch — The real AI race may no longer be at the frontier → · Quartz — Moonshot AI releases Kimi K3 open-weight model for download → · NIST — CAISI assessment of Z.ai's GLM-5.2 →