Over Half of LLM Requests Go to Models Younger Than 60 Days, Vercel Gateway Data Shows
aruthesa · reddit · 2026-11-16
Citing Vercel's latest AI Gateway report (October data): more than half of language-model requests go to models less than 60 days old, and about a month after a model's peak usage, roughly three in four teams that used it at peak have already moved on (Vercel gateway traffic only, not the whole industry).
The author opens a discussion on how teams decide when to switch models in agent workflows: what to compare (quality on own tasks, cost, latency, tool-call reliability, rate limits), fixed eval sets vs. vibes, switching frequency, models that looked better on paper but broke tool calls, and whether people use a single gateway/router (they currently use io intelligence) to swap models without touching integration code.
More from coding & agent
- I Shipped an App, a Site, and a CRM With Codex and Claude Code — and Learned Nothing — ThatGuyJord2020 · 2026-11-16
- Craft Anything is early alpha: no login, DRM or anti-cheat bypass — thisguyknowsai · 2026-11-16
- Open-source Craft Anything hits 34K GitHub stars in 16 days by letting Claude Code clone any app — thisguyknowsai · 2026-11-16
- Loop, Harness, Graph: What Each Agent Layer Fixes and What It Breaks — Full-Presence7590 · 2026-11-16
- Agents Added 64 Type Assertions in 3 Months to Silence TypeScript Errors Instead of Fixing Them — Puzzleheaded-Fox9821 · 2026-11-16
- Agent discovers local Blender install and uses it to autonomously produce striking 3D graphics — n_sri_laasya · 2026-11-16