GPT-5.5 arrived just six weeks after the previous model in the same line, an interval that describes the sector's pace in the period well.
Six weeks is shorter than the planning cycle of most companies building products on these models. When the next generation arrives before the previous one is fully integrated, teams have to decide whether to migrate or accumulate lag.
Neither choice is comfortable. Migrating at every launch consumes development time on work that produces no new features; not migrating means paying more for worse results as the older generation's pricing fails to keep pace.
The pace also affects quality assessment. Careful evaluation takes weeks, and a model swapped before then means the organisation never quite learns what it was using.
The practice that survives that pace is decoupling: keep model choice in a configuration layer, with automated tests over the operation's real task, so switching becomes a measurement decision rather than a migration project.