Agent Infra Benchmarks
Independent measurements of the infrastructure agents run on. Same method for every vendor, raw data for every number, and failures shown as failures.
Task 1Live
Sandbox time to first command
Cold create to the first successful shell command, on each provider's default image. 5 providers measured, 6 listed with the reason they could not be. Current p50 leader: Cloudflare Sandbox.
SuggestOpen
Want a provider or a task covered?
Tell me which sandbox, agent VM or browser provider to add, or which task to measure next. DM on X.