What one build workload — a cold clone + install + typecheck of the opencode repo (the Dax benchmark) on an 8 vCPU / 16 GiB sandbox — actually costs on each provider, using measured workload times and published rates. Slower providers pay twice: once in latency, once in billable seconds.
Benchmark run: September 18Configured duration x each provider's published rate at the chosen spec. Sorted by cost, lowest first.
Workload
Dax build workload
Cold clone + install + typecheck of opencode on a fresh sandbox — the workload we benchmark daily. Move the sliders to price your own shape.
Bars are scaled to the most expensive provider. Bar length = estimated monthly cost at 1,000 workloads of a 8 vCPU / 16 GiB sandbox at 60s each. Rates recompute from per-unit pricing where published; "flat" providers bill per sandbox regardless of spec. Idle billing varies: wall-clock providers charge in full, active-CPU providers drop to memory-only rates, and Sail/Givemeanode sleep to ~zero. Sources below.
$0.0000375/core-s (1 core = 2 vCPU, so 4 cores) + $0.0000064/GiB-s.
Source pricing page$0.0000115/GB-s on allocated RAM while active (16 GiB). CPU is derived from memory — 16 GB maps to ~6 cores, not 8. Upper bound shown.
Source pricing page$0.0504/vCPU-hr + $0.0162/GiB-hr — parity with E2B. No egress fees.
Source pricing page$0.000014/vCPU-s + $0.0000045/GiB-s — identical rates to E2B. Creating/stopping transitions bill as running.
Source pricing pageRollout sandboxes at $0.04572/GiB-hr while active (16 GiB); parked/idle drops to ~$0.000139/GiB-hr. Interactive cpu-8 node is $0.4824/hr with a 15-min minimum.
Source pricing page$0.025/vCPU-hr + $0.015/GiB-hr. Published profile for 8 vCPU + 16 GiB is exactly $0.44/hr.
Source pricing pageSandbox rate $0.00003942/core-s (4 cores = 8 vCPU) + $0.00000667/GiB-s. Regional multipliers (1.25–2.5x) not included.
Source pricing page8 vCPU/16 GB shape = 8 unit-min/min at the pay-as-you-go rate of $0.0015/unit-min. Team/Business plans include prepaid minutes at an effective $0.001/unit-min ($0.48/hr). Bills in 1-minute minimums with 15-second rounding.
Source pricing page$0.108/CPU-hr + $0.0252/GB-hr, billed on top of a plan.
Source pricing page$0.015/used vCPU-hr + $0.008/used GiB-hr + $0.0007/GiB-hr NVMe — bills observed usage, not reservations (upper bound shown at full utilization). Plus $0.012 per L Sailbox creation. No charge while sleeping.
Source pricing page$0.02/GiB-hr on the memory limit (16 GiB). Paused sandboxes stop compute charges.
Source pricing page$0.0504/vCPU-hr + $0.0162/GiB-hr + storage $0.000108/GiB-hr.
Source pricing page$0.000014/vCPU-s + $0.0000045/GiB-s; usage draws from monthly workspace credits first.
Source pricing page$0.07/core-hr active CPU + $0.015/GB-hr RAM (on-demand rates; Pro is ~40% lower).
Source pricing pageLarge box (8 vCPU / 16 GB): $0.40/active-CPU-hour — assumes 100% CPU utilization for the whole build.
Source pricing page$0.128/active-CPU-hr per vCPU + $0.0212/GB-hr provisioned memory (wall clock). CPU upper bound at 100% utilization; real bills for I/O-bound runs are lower.
Source pricing pageEach provider runs the same workload in a fresh 8 vCPU / 16 GiB sandbox: prepare, Bun download/unpack, shallow clone of opencode, bun install, and bun typecheck. Cost per workload = configured duration x the provider's published rate at the chosen spec.
Caveats: workload time excludes sandbox create/destroy, which most providers also bill. Providers on "active CPU" billing (Vercel, Modal, Blaxel, Cloudflare, Upstash, Tensorlake) may charge less than the wall-clock figure shown — treat those as upper bounds; the Dax workload is compute-heavy, so the gap is smaller than for I/O-bound agent workloads. Rates verified September 18, 2026 against each provider's public pricing page; subscription floors and included credits are not modeled.