Billing, limits & error reference

Browser fleets reuse the same billing, quota, and lifecycle machinery as browser tests — there is no separate fleet plan or new meter to learn.

How fleets are billed

A fleet is metered exactly like a browser load test: one running browser = one browser-class virtual user, per location, per minute, drawn from your plan’s VU-hours.

• You are billed for browsers that are allocated and running, per location, for as long as the fleet holds them — regardless of whether a session is actively driving a browser at that moment.
• Metering starts when a browser becomes ready and stops when it is released (fleet deleted, idle timeout, or TTL reached).
• Fleet minutes and browser-test minutes both draw from the same VU-hours pool — a fleet does not add a new line item, and it is never double-counted against a browser test.

Stop paying for idle browsers

Two controls keep spend down: set a short idle timeout so an unused fleet releases itself, and a TTL so no fleet outlives its job. Or delete_browser_fleet / DELETE /v1/fleets/<fleetId> the moment you’re done.

Packing: browsersPerRunner

browsersPerRunner controls how many browsers share a runner — the fleet analogue of “virtual users per runner” on test create. It does not change what you’re billed (that’s per browser, per location); it changes video layout and runner count:

• 1 → each session runs solo with a full-screen video.
• > 1 → browsers are tiled on a runner, each with its own tile view.

It defaults to your plan’s browser policy and is clamped to your tier’s minimum and maximum, just like test create.

Limits & timeouts

Limit	Default	Notes
Session idle timeout	300 s	A single WebDriver / Playwright session with no activity is closed. Start a new session to continue.
Fleet idle timeout	15 min	A fleet with no active sessions releases all its browsers and stops billing. Set per fleet via idleTimeoutMinutes.
Fleet TTL (lifetime)	120 min	Hard cap on how long a fleet lives, active or not. Set per fleet via ttlMinutes.
Maximum TTL	480 min	Ceiling on ttlMinutes; requests above it are clamped.
Browser counts / concurrency	plan-dependent	Total browsers are gated by your plan’s browser-VU quota, same as browser tests.

Note

Timeouts are enforced by the fleet’s supervising workflow, not the connection — so an abandoned fleet always tears itself down and never leaks running browsers.

Error reference

The gateway returns protocol-appropriate errors so your existing client handles them naturally.

Situation	Response
Missing or invalid credentials (WebDriver)	401 Unauthorized with a WWW-Authenticate: Basic challenge
Missing or invalid token (Playwright / CDP)	WebSocket connection closed / rejected
Playwright client minor not supported	HTTP 428 with a body listing supportedPlaywrightVersions
Unknown or already-ended session	W3C invalid session id error
No free browser after the queue wait (~60 s)	W3C session not created error
Fleet has ended (TTL, idle, or deleted)	410 Gone on every endpoint
Plan / quota limit hit on create	402 Payment Required

Match your Playwright version

The 428 is the most common fleet error. Playwright’s connect() requires your local @playwright/test minor to match a fleet-supported minor — read the returned supportedPlaywrightVersions and pin to one, or use the CDP escape hatch for a version-agnostic Chromium connection.

See also

• Browser fleets overview — endpoints, auth, and the capability matrix.
• Connect a client — protocol-agnostic endpoint and credential reference.
• Playwright quickstart — version matching and CDP.
• Selenium quickstart — Python / Java / JS and the REST create call.
• Billing — how VU-hours and plan tiers work across MaxoPerf.