Interactive tool calls
HOW THIS IS RANKED
Time from requesting a sandbox until its first command succeeds. This is the median of a burst of 100 starts. Lower is better.
ComputeSDK · Nov 6, 2026 ↗One API.
Every sandbox.
Run workloads across sandbox providers through a single integration. Optimize for cost, performance, and availability without vendor lock-in.
TRIGGERSPROVIDERS
- GitHub
- Linear
- Webhook
- API & CLI
router
- Provider Xout of capacity
- Daytonarunning
- E2B
- Standby Pool
sandboxrouter runs events run_8f2k --follow
- 14:02:07.081attempt.startedgithub pull_request.opened
- 14:02:07.094execution.placing[provider-x, daytona, e2b]
- 14:02:07.233 ↳ provider-xout of capacity → next
- 14:02:08.517execution.startedcurrentProvider=daytona ✓
- Sandboxes
- 13.3K
- Failover
- 0
- Providers
- 5
Find the right sandbox for the job.
Compare providers on the benchmark that matters for your workload.
Parallel evals and fan-out
HOW THIS IS RANKED
99th-percentile time to interactive in the same burst of 100 sandboxes. Fan-out workloads depend on the slowest starts. Lower is better.
ComputeSDK · Nov 6, 2026 ↗Coding agents and CI
HOW THIS IS RANKED
Time to prepare a fresh sandbox, clone OpenCode, install dependencies and typecheck it in ComputeSDK’s DAX benchmark. Lower is better.
ComputeSDK · Nov 6, 2026 ↗Long-running agents
HOW THIS IS RANKED
Catalog providers with memory snapshots and fork, ordered by ComputeSDK’s start score. The score combines median, p95 and p99 start times with success rate. Higher is better; this is not a persistence benchmark.
ComputeSDK · Nov 6, 2026 ↗GPU workloads
HOW THIS IS RANKED
GPU-capable providers in our catalog, ordered by their general sandbox start score. This measures neither GPU throughput nor GPU provisioning time. Higher is better.
ComputeSDK · Nov 6, 2026 ↗Batch and background jobs
HOW THIS IS RANKED
The cost of one clone, install and typecheck workload, as published alongside ComputeSDK’s DAX benchmark. Lower is better. It is not a quote for your workload.
ComputeSDK · Nov 6, 2026 ↗Latency, uptime, cost. Optimize all three.
Compare the trade-offs, choose your provider order, and let Sandbox Router handle placement and failover.
Compare providersYour routing orderIllustration
Keep running through provider limits.
When a provider can’t start a sandbox, Auto tries the next eligible provider. The Standby Pool is there when the providers in your order can’t take the run.
How failover worksProvider timelineFailover example
One dashboard for every provider
Keep sandbox activity and provider usage in view from one console, without switching between dashboards.
See the consoleUsage consoleExample workspace
Runs12,403
Failover318
Spend$170.35
14 days agoToday
ProviderRunsSpend
Daytona5,752$63.27
E2B3,554$53.31
Cloudflare2,107$37.93
Other990$15.84
Control where your code runs
Compare isolation and security requirements, then choose the providers that fit your workload.
Explore provider capabilitiesRouting policyIllustrative requirements
Require
ProvidermicroVMSOC 2Confidential ComputeResult
Provider A✓––matches
Provider B✓✓–matches
Provider C✓✓–matches
Provider D✓✓–matches
Provider E–––matches
Provider F✓✓✓matches
Route your first run.
One API for every sandbox provider.