Vercel Flags cuts cold-start latency 7x, stops billing for usage reports
A rework of the Flags SDK removes a network wait from flag initialisation and moves usage reporting off the request path, so it no longer extends function duration.
Vercel reworked its Flags SDK on 29 October to fix two separate costs that were hitting every flag check. The SDK now loads and parses the flags datafile bundled with your deployment as soon as the client is created, instead of waiting on a network round trip to confirm it. Vercel says this cut p75 cold-start initialisation time from over 70ms to 10ms, roughly seven times faster, with p99 flag-read latency now under 1ms across cold and warm invocations.
The second fix targets billing, not speed. The SDK previously used waitUntil to report usage after a response was sent, which kept the function running longer than the request needed. Usage events now go to the Vercel runtime directly, so reporting no longer adds to function duration.
pnpm add @flags-sdk/vercel@latestUpdate @flags-sdk/vercel to 1.5.3 or later and redeploy to get both changes.
Why it matters: if you’re evaluating flags on every request, as most teams do, both changes land directly on your bill and your latency numbers without any code change beyond a dependency bump. The usage-reporting fix in particular was invisible until now: function duration padding from waitUntil doesn’t show up unless you go looking for it.