Vercel Data Shows AI Models Have Six-Week Half-Life in Production

Why it matters

Vercel's November 2026 AI Gateway Production Index finds that AI models in production have an effective half-life of roughly six weeks, with the majority of October's language-model requests going to models under two months old. Because Vercel's gateway routes enormous token volume between production applications and model providers, its data reflects real enterprise behavior rather than lab benchmarks. The report shows teams abandoning models within about a month of peak usage, often switching to different labs rather than staying within one family. It also documents the rise of decision models, which classify or score rather than generate text and now handle a quarter of gateway requests at a tiny fraction of the spend, plus accelerating voice adoption and record image and video spending. The competitive picture is shifting too: Anthropic now wins new teams at nearly twice OpenAI's rate, while OpenAI captures a growing share of total spend. The core implication is that providers cannot rely on lock-in, and enterprises must build pipelines that tolerate constant model swapping.