coldtally

Startup times

How long a new sandbox or cloud browser takes to become usable, from ColdTally's daily benchmark: 10 timed runs per provider each day. A secondary check next to price, not a speed leaderboard. Sandbox and browser numbers time different steps, so compare within a board only.

Measured 2026-10-18 Sandboxes Cloud browsers How we measure Cost per task

Sandboxes

4 of 6 measured12 · fastest median 225 ms (E2B)
Conditions
Timed
create + first command3
Samples
1 warm-up + 10
Runner
GitHub Actions, Phoenix, US
Regions
not all pinned4
How we measure
Provider Median, ms p95, ms Size5 Region Runs ok
Cloudflare Sandbox6 169 ms 279 ms standard-1 (1/2 vCPU, 4 GiB) LAX4 10/10
1 E2B 225 ms 668 ms default default4 10/10
2 Modal 567 ms 1,554 ms 1 core (2 vCPU), 4 GiB default4 10/10
3 Daytona 1,207 ms 1,561 ms default us 10/10
Measured 4 of 6 measured · Raw results
Notes 6
  1. Not yet measured: Vercel Sandbox. Help us measure Vercel Sandbox
  2. Pricing only. Blaxel's terms prohibit publishing benchmarks or performance information about Blaxel (Terms & Conditions, restriction (f)).
  3. Timed from the create call until the command echo ok returns in the new sandbox (create + first command). Destroy time is recorded separately. Sandbox and browser boards time different steps, so their numbers are not comparable with each other.
  4. Region not pinned: the provider chose placement, because region selection is not offered, costs extra, or needs a paid plan on that provider. Where the provider reports a region, the observed one is shown.
  5. Instance sizes differ: each provider runs at the size listed in the Size column (provider default unless stated). Startup time can depend on size and image.
  6. Cloudflare Sandbox is timed inside its benchmark Worker, within Cloudflare's network, so it excludes the network latency from the runner that the other rows include. The full request from the runner (network, Worker routing, start, echo ok and destroy) took a median 746 ms.

Median time to a usable sandbox in ColdTally's latest daily run. Grey rows are timed inside the provider's network, so they take no rank. Prices are the main comparison. Method. Ask your agent.

Cloud browsers

4 of 6 measured1 · fastest median 882 ms (KERNEL)
Conditions
Timed
create + connect + page load2
Samples
1 warm-up + 10
Runner
GitHub Actions, Phoenix, US
Target
coldtally.com/bench-target/
Regions
not all pinned3
How we measure
Provider Median, ms p95, ms Region Runs ok
Cloudflare Browser Run4 666 ms 1,363 ms LAX3 10/10
1 KERNEL 882 ms 1,036 ms us-east3 10/10
2 Browser Use 1,135 ms 2,286 ms default3 10/10
3 Steel 1,650 ms 2,532 ms iad 10/10
Measured 4 of 6 measured · Raw results
Notes 4
  1. Not yet measured: Browserbase, Hyperbrowser. Help us measure Browserbase · Help us measure Hyperbrowser
  2. Timed from the create call until the target page reaches DOMContentLoaded (create session + CDP connect + page load). Release time is recorded separately. Sandbox and browser boards time different steps, so their numbers are not comparable with each other.
  3. Region not pinned: the provider chose placement, because region selection is not offered, costs extra, or needs a paid plan on that provider. Where the provider reports a region, the observed one is shown.
  4. Cloudflare Browser Run is timed inside its benchmark Worker, within Cloudflare's network, so it excludes the network latency from the runner that the other rows include. The full request from the runner (network, Worker routing, launch, connect, page load and release) took a median 777 ms.

Median time to a usable browser in ColdTally's latest daily run. Grey rows are timed inside the provider's network, so they take no rank. Prices are the main comparison. Method. Ask your agent.