---
type: "post"
id: "vercel-sandbox-now-starts-up-to-4x-faster-b2aaf9"
slug: "vercel-sandbox-now-starts-up-to-4x-faster-b2aaf9"
title: "Vercel Sandbox now starts up to 4x faster."
description: "Vercel says Sandbox initialization can be up to four times faster outside Washington, D.C., when a regional cache serves reused image data."
retrieval_nugget: "Vercel says Sandbox initialization can be up to four times faster outside Washington, D.C., when a regional cache serves reused image data. Its internal tests found the largest cited gains in Paris and Seoul. For teams that repeatedly create remote environments, this can reduce startup waiting time without a stated configuration change."
published_at: "2026-11-13"
updated_at: "2026-11-13"
record_date: "2026-11-13"
date_kind: "published_at"
topics: ["sandbox"]
entities: ["Vercel","Vercel Sandbox","Vercel Container Registry"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "Vercel reports a separate Sandbox startup improvement of up to 4x outside iad1 through regional caching of reused image data, with benchmarks showing roughly 75% faster starts in Paris and Seoul."
prior_refs: ["https://newruntime.com/posts/vercel-sandboxes-with-drives-start-2-4x-faster-4873e5/","https://newruntime.com/posts/sandbox-snapshots-are-now-up-to-10x-faster-1bd269/"]
source_url: "https://vercel.com/changelog/vercel-sandbox-starts-up-to-four-times-faster"
source_urls: ["https://vercel.com/changelog/vercel-sandbox-starts-up-to-four-times-faster"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:vercel-sandbox-now-starts-up-to-4x-faster-b2aaf9"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/vercel-sandbox-now-starts-up-to-4x-faster-b2aaf9/","markdown":"https://newruntime.com/posts/vercel-sandbox-now-starts-up-to-4x-faster-b2aaf9.md","json":"https://newruntime.com/posts/vercel-sandbox-now-starts-up-to-4x-faster-b2aaf9.json"}
markdown_url: "https://newruntime.com/posts/vercel-sandbox-now-starts-up-to-4x-faster-b2aaf9.md"
---

# Vercel Sandbox now starts up to 4x faster.

## Retrieval answer

Vercel says Sandbox initialization can be up to four times faster outside Washington, D.C., when a regional cache serves reused image data. Its internal tests found the largest cited gains in Paris and Seoul. For teams that repeatedly create remote environments, this can reduce startup waiting time without a stated configuration change.

## Summary
Vercel says Sandbox initialization can be up to four times faster outside Washington, D.C., when a regional cache serves reused image data. Its internal tests found the largest cited gains in Paris and Seoul. For teams that repeatedly create remote environments, this can reduce startup waiting time without a stated configuration change.

## What happened
Vercel’s changelog says Vercel Sandbox now initializes up to four times faster outside the Washington, D.C. region, identified as `iad1`, when reused image data is served from a regional cache. The update changes where Sandbox initialization reads image data. On a cache hit, the system reads cached data close to the region in which the sandbox runs, rather than making a round trip to origin image storage in `iad1`.

That distinction matters because image retrieval is part of the path before a fresh environment is ready. A sandbox in a remote region previously had a longer network journey to the origin store for relevant image data. Vercel says the regional cache avoids that trip on cache hits. The company frames the result as faster initialization, rather than as a new Sandbox execution mode or a change in how teams define sandbox workloads.

According to the announcement, regional caching works with managed images and custom images backed by Vercel Container Registry. It also applies to snapshots based on those images. That scope includes the image and snapshot forms used to create Sandbox environments, but Vercel does not describe a separate opt-in control, migration procedure, or required image rebuild. The changelog links to the Vercel Sandbox documentation for further product details.

Vercel reports its measurements from internal end-to-end benchmarks using Vercel Managed Images. In Paris, startup time fell by about 75%. In Seoul, the company says startup time declined from about 3.5 seconds to 1.05 seconds. The company also says regions nearer to the `iad1` origin store generally see smaller improvements, since their requests already travel a shorter distance.

The “up to four times faster” figure is therefore conditional rather than a promised result for every Sandbox run. It applies outside `iad1` where reused image data is served from a regional cache. The changelog does not list results for every region, distinguish image sizes, or state how often workloads receive cache hits. It also does not announce a pricing change, a revised Sandbox limit, or a change in the underlying execution API. The stated change is specifically the location from which cached image data is read during initialization.

For an engineering team, the immediate question is not whether every Sandbox becomes four times faster. It is whether its own environment creation path is remote from `iad1`, based on reusable images, and frequently enough repeated that a shorter startup becomes visible in the workflow. The Paris and Seoul examples identify the kind of geography where Vercel says the difference can be substantial. They do not establish equivalent gains for every Vercel region or customer image.

## What the archive adds
New Runtime’s earlier coverage on [Drive-enabled Sandbox startup](https://newruntime.com/posts/vercel-sandboxes-with-drives-start-2-4x-faster-4873e5/) (November 2, 2026) recorded Vercel’s report of up to 2.4× faster startup, including median and p95 measurements. A later note on [Sandbox snapshots](https://newruntime.com/posts/sandbox-snapshots-are-now-up-to-10x-faster-1bd269/) (November 10, 2026) covered faster snapshot creation and restoration. This is a new performance fact in an ongoing story about reducing Sandbox startup and snapshot latency, not a restatement of either earlier change.

## Trend signal
This update does not yet fit a pattern New Runtime tracks. It is a specific infrastructure performance improvement, and Vercel’s announcement does not establish a broader change in agent architecture, economics, or workflow. The earlier Sandbox updates show repeated optimization of environment startup and snapshots, but that alone is not enough to claim a wider trend.

## Why it matters
For teams that create short-lived environments, startup time is the interval before a developer, automated workflow, or agent can begin doing useful work. When a Sandbox runs far from `iad1`, image retrieval can be a meaningful part of that interval. Vercel’s regional-cache approach targets this portion of the path rather than changing the code or command that runs inside the environment.

The likely practical benefit is greatest in workflows that recreate similar environments repeatedly. A team may launch isolated Sandboxes for task execution, testing, or remote development work; when the same image data can be reused from a nearby cache, the wait before each environment is ready may fall. The Seoul example is the most concrete illustration in the announcement: Vercel reports a reduction from roughly 3.5 seconds to 1.05 seconds.

That benefit should not be confused with an end-to-end productivity or cost claim. Faster initialization does not establish that a complete agent task, test suite, or deployment is faster. Nor does the update by itself say anything about concurrency, reliability, execution duration, or image management costs. It is a narrower optimization of the environment bootstrap path. Teams should evaluate it against their own regional placement, image reuse, and actual cache behavior.

## What happens next
- **Vercel will show further startup-latency improvements around image reuse or snapshot restoration.** (6 months, 60%). Watch for: regional benchmark figures or additional cache coverage.
- **Teams with Sandbox workloads far from `iad1` will measure materially different gains by region and image reuse.** (3 months, 70%). Watch for: customer measurements comparing cold and cached initialization.

## Limits and uncertainty
The reported figures are Vercel’s internal end-to-end benchmarks with Managed Images. The company does not publish an independent benchmark, full test methodology, sample size, variance, cache-hit rate, or image-size breakdown. Paris and Seoul are examples, not a per-region performance table. The “up to four times” statement depends on a regional cache hit and should not be interpreted as an expected multiplier for every initialization.

Vercel also does not quantify the update’s effect on total workflow duration, task completion time, or cost. The changelog says managed and custom Container Registry-backed images, plus snapshots based on them, are supported, but it does not provide operational detail about cache population, cache lifetime, or behavior during a cache miss. Those factors determine how much of a team’s traffic can receive the reported benefit.

## Skill to build now
Instrument Sandbox creation separately from the work that follows it. This week, record initialization time by execution region and image identity, then compare repeated launches with first launches where possible. Keep the measurement distinct from command runtime or test duration: Vercel’s stated improvement concerns cached image reads during initialization. If your workload uses reusable Container Registry-backed images in remote regions, this gives a direct way to see whether the announced path affects the actual bottleneck.

## Reader takeaway
Vercel is moving cached Sandbox image reads closer to the execution region, with its largest reported gains far from `iad1`. The change can reduce environment startup latency for reusable images, but it is conditional on cache hits and geography. Measure initialization in your own regions before turning Vercel’s benchmark into a workflow or cost assumption.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Sandbox](https://newruntime.com/topics/sandbox.md): Explore the sandbox topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)