AI Model Half-Life Falls to Six Weeks, Vercel Data Shows

What happened

Vercel's data indicates that a typical AI model reaches its peak within a month of launch, and that the half-life of model usage is roughly six weeks. TechRadar reported that most AI traffic on the Vercel platform now runs on models less than two months old.

The findings point to development teams moving to newer models at a pace close to the speed at which AI labs release them. Loyalty to any single model is therefore becoming less durable, and the window in which a model holds peak relevance is short.

Why it matters

For organisations building on AI, model choice is no longer a one-off procurement decision. If usage concentrates on recent releases, teams need architectures that can swap models without rebuilding the product around them. Integration layers, evaluation routines and prompt libraries become the durable assets, while the underlying model becomes more interchangeable.

The same dynamic affects how products are experienced. Users who notice a change in tone, speed or capability after a model switch will judge the service on that change, so version transitions need to be managed as visible product moments rather than quiet back-end swaps.

By the numbers

• Around one month: the typical period in which a newly launched AI model peaks, according to Vercel's data as reported by TechRadar.
• Six weeks: the half-life Vercel's data attributes to AI models.
• Two months: the age threshold below which most AI traffic on the platform is running.

The Renascence take

Most readers will focus on the speed of model turnover and overlook what it reveals about user behaviour. People are not loyal to a model; they are loyal to outcomes that feel reliable, and they will move to whatever seems to deliver them next.

Treat model selection as a service-design decision, not a procurement one. Build experiences that hold their shape when the engine changes, and measure satisfaction across each switch rather than assuming the last release was the right one. If your product depends on a single model staying the same, the next six weeks will expose that dependency.