E2B316 ms/02
blaxel598 ms/03
Tensorlake895 ms/04
declaw1.05 s/05
Vercel1.23 s/06
Daytona1.31 s/07
CodeSandbox2.93 s/08
Dedalus4.96 s/09
Sprites6.82 s/10
CLOUDFLARE—/11
hopx—/12
Modal—/13
Northflank—/14
RUNLOOP—/The live race
for cloud sandboxes.
- Pick a workload.
- ABTwo sandboxes run the same trial side by side —
- Streaming logs, percentiles, real $/run.
- The evidence decides who takes the flag.
Arena circuit · sector 03
Cloud sandbox grand prix
- On circuit
- 08
- Full field
- 14
- Finish
- Evidence
A benchmark workspace
From question to reproducible trial without leaving the page.
The racing metaphor ends at the hero. The product underneath is a quiet, configurable instrument.
Build a trial
Choose a catalog workload or paste shell commands, set resources and repeat the same run.
Open workspace02Read the evidence
Inspect median, p95, p99, success, cost and the sample behind every rank.
View benchmarks03Narrow the field
Select 2–4 providers and compare product constraints beside measured results.
Compare providersProvider directory
Fourteen runtimes. One trial contract.
Current evidence
A rank is only useful when the uncertainty travels with it.
Every benchmark view keeps sample size, tail latency, failures and cost beside the headline result.
- Measured runs
- 744
- Fastest median
- 895 ms
- Fastest p95
- —
9 providers tested
Tensorlake
Awaiting data
Configure
Providers, commands, resources, region and repetitions are explicit.
Bring credentials
Use session-only API keys so provider usage lands on your own account.
Take it with you
Download the complete trial config and export local result history.
Start with the audited catalog. Paste your own workload when you are ready to spend your own provider credits.
Configure a trial