Benchmarks
Nimbus creates a session and runs the first command in 0.74 s median. The tables below give the method, the date, and the spread.
Session create → first command
Section titled “Session create → first command”This table uses the ComputeSDK method, run on 2026-08-08. Each result is
wall-clock sandbox.create() plus the first successful
runCommand('node -v'), 5% trimmed. The composite weighs normalized scores
of the median (0.60), P95 (0.25), and P99 (0.15) on a 10-second scale, times
the success rate.
| Mode | N | Success | Median | P95 | P99 | Composite |
|---|---|---|---|---|---|---|
| Sequential | 200 | 200/200 | 0.74 s | 1.00 s | 1.06 s | 91.5 |
| Staggered (200 ms) | 200 | 200/200 | 0.75 s | 1.02 s | 1.08 s | 91.3 |
| Burst (simultaneous) | 200 | 200/200 | 1.02 s | 1.41 s | 1.49 s | 88.1 |
These are per-request Durable Object boots with no warm pool. Each mode is
two runs of 100, pooled, so the table covers 600 sandboxes with no failures.
create() took a 701 ms median and the first command 35 ms, so session boot
is ~95% of TTI. After that a command round-trip takes ~31 ms.
The client was a residential Linux machine and the target a workers.dev
deployment of this repo’s probe embedder. Network RTT to the edge is inside
these numbers. A no-op RPC to an already-booted session costs 31 ms of
them. The 2026-07-16 run of this table reported 0.61 s sequential. That
same build, re-measured alongside this one on 2026-08-08, gives 0.74 s
from this client. The two dates are separate measurements rather than a
series.
Nimbus is not on the official ComputeSDK leaderboard; the unmodified harness ran locally. Treat cross-provider numbers as directional. Against Cloudflare’s container-based Sandbox SDK, Nimbus measured 2.6× faster on the sequential median, 0.74 s against 1.95 s. On burst it measured 4.0× faster, 1.02 s against 4.05 s. Those Sandbox SDK figures come from its published leaderboard data of 2026-07-13.
Everyday operations
Section titled “Everyday operations”Measured against the live deployment in 2026-07.
| Operation | Wall time |
|---|---|
git clone facebook/react (7,300 files) |
~28 s |
npm install 611-package app (52,500 files, cold) |
~165 s |
npm install zod (cold session) |
~6 s |
node -e 'console.log(...)' (warm) |
102–152 ms |
| Vite hot reload | 302 ms median |
Cross-session caching gives 9–16× warm-vs-cold speedups on package installs.
tests/behavioral/perf-regression/ checks these thresholds on every
deploy. Run it against any deployment:
BASE=https://nimbus-os.dev bun test:behavioral