How do cloud browsers bill session seconds, proxy megabytes, and CAPTCHA attempts?

One cloud-browser hour can mean 120 Browserless Units, 100 Hyperbrowser credits, $0.09 plus bandwidth at ZenRows, or no time charge at all because Bright Data bills only transferred gigabytes. Browserless, Hyperbrowser, ZenRows, and Bright Data, retrieved 2026-07-26.

The answer is to price each independent meter, then apply the plan allowance and minimum purchase: browser time, proxy traffic, CAPTCHA attempts, initialization, requests or steps, and sometimes concurrent capacity can all appear on the same bill. Browserless, Browser Use, and Cloudflare, retrieved 2026-07-26.

The billing formula that survives vendor terminology

For a session-based product, use this shape and delete the terms that the vendor does not charge:

billable usage
  = rounded session time
  + proxy traffic
  + CAPTCHA attempts
  + browser creation or task initialization
  + API calls, agent steps, or model tokens
  + concurrent-capacity charges

cash due
  = plan minimum
  + usage above each included allowance

Do not add every published figure blindly. Browserless explicitly adds proxy and CAPTCHA Units to browser-time Units, while Browserbase separately meters browser hours and proxy gigabytes. By contrast, Bright Data Browser API replaces the whole time-and-request model with bandwidth billing. Browserless unit consumption, Browserbase pricing, and Bright Data Browser API, retrieved 2026-07-26.

Every meter in the current cloud-browser vendor records

The table includes every product tagged cloud-browser in the current dataset. Some are remote-browser sessions. Others are request APIs or platforms that expose a browser-backed path, so an hourly comparison is not always possible.

Every meter in the current cloud-browser vendor records table

Why a per-page estimate gives the wrong answer

A page is an application event. A billable unit is a vendor event. Those events are not interchangeable.

Browser Use Cloud makes the billing unit choice explicit: browser sessions bill time and proxy data, while an agent task can add initialization and V2 step charges. The worked example below prices all four meters without treating browser time as a per-page fee. Browser Use pricing, retrieved 2026-07-26.

Bright Data creates the opposite problem. It has no session-time charge, so two one-hour jobs can have different bills solely because one transfers more bytes. ZenRows charges both dimensions, while Cloudflare can add monthly-average concurrency to time for direct sessions. Bright Data, ZenRows, and Cloudflare, retrieved 2026-07-26.

Request-backed products need yet another billing unit. ScrapingFish charges successful responses, Zyte charges successful browser-rendered responses by difficulty tier, and ScrapeNinja consumes a JavaScript request quota plus possible bandwidth overage. “One hour” does not reveal how many requests succeeded, which target tier applied, or how many MB crossed the wire. ScrapingFish, Zyte, and ScrapeNinja, retrieved 2026-07-26.

One-hour meter reference across vendors

This comparison uses one browser for exactly 3,600 active seconds. It excludes proxy traffic, CAPTCHA solving, API calls, AI work, storage, and concurrency above the plan allowance. “Included” means the hour consumes an existing allowance; it does not mean the plan itself has no purchase price.

One-hour meter reference across vendors table

The one-hour table deliberately does not turn plan credits into an invented universal dollar rate. A plan price buys capacity, allowances, retention, and concurrency as well as usage. Dividing monthly price by one allowance can look precise while mispricing every other meter in the same shared pool. Browserless pricing and Browserbase pricing, retrieved 2026-07-26.

Fully worked example: Browser Use Cloud session and task meters

Assume a Pay As You Go job keeps one browser open for exactly 5,400 seconds, or 1.5 hours, transfers exactly 0.25 GB through the residential proxy, starts one task, and uses 20 V2 agent steps. Browser Use Cloud charges $0.06 per browser hour, $10 per proxy GB, $0.01 per task initialization, and $0.002 per V2 step on Pay As You Go. Browser Use pricing, retrieved 2026-07-26.

browser time = 1.5 hours × $0.06 per hour = $0.09
proxy data   = 0.25 GB × $10 per GB       = $2.50
task start   = 1 task × $0.01             = $0.01
V2 steps     = 20 steps × $0.002          = $0.04
total usage                                   $2.64

Pay As You Go requires prepaid credit purchases of $25, $50, $100, or $500. One job therefore consumes $2.64 of the minimum $25 purchase and leaves $22.36. Ten identical jobs consume $26.40, which is $1.40 more than one $25 pack; the minimum published packs needed total $50 and leave $23.60. Browser Use pricing, retrieved 2026-07-26.

1 job                  =  1 × $2.64 = $2.64
balance after 1 job    = $25 - $2.64 = $22.36
10 jobs                = 10 × $2.64 = $26.40
shortfall after $25    = $26.40 - $25 = $1.40
balance after $50      = $50 - $26.40 = $23.60

These totals do not estimate failure cost because Browser Use Cloud does not publish whether a failed task reverses initialization or step charges. Browser Use pricing, retrieved 2026-07-26.

Fully worked example: Browserless CAPTCHA attempts expose a source conflict

Assume one hypothetical Browserless job uses 60 browser seconds, 2 residential proxy MB, and 2 CAPTCHA attempts, with one failed attempt followed by one successful attempt. Browserless’s Unit documentation records 1 Unit per started 30 seconds, 6 Units per residential MB, and 10 Units per CAPTCHA attempt whether they succeed or fail. Browserless Unit consumption, retrieved 2026-07-26.

session time = 60 seconds / 30 seconds per Unit = 2 Units
proxy data = 2 MB x 6 Units per MB = 12 Units
CAPTCHA by the Unit documentation = 2 attempts x 10 Units per attempt = 20 Units
documented-attempt total = 2 + 12 + 20 = 34 Units

The pricing page uses different wording and lists 10 Units per successful CAPTCHA solve. Under that narrower description, only the one successful solve in the same scenario is counted:

CAPTCHA by the pricing-page wording = 1 successful solve x 10 Units = 10 Units
successful-solve total = 2 + 12 + 10 = 24 Units

The public sources therefore support a range of 24 to 34 Units for this exact scenario, not one verified total. The dataset stores the attempt-based 10-Unit meter because that is the explicit billing unit in the Unit documentation, but the pricing-page conflict means failed-attempt billing cannot be stated as settled. Verify the exported Unit log or invoice before forecasting repeated failed solves. Browserless Unit consumption and Browserless pricing, retrieved 2026-07-26.

What an idle or abandoned session can cost

“Idle” is vendor-specific. Browserless bills an open idle connection exactly like an active one. Kernel enters zero-cost standby after five seconds without a CDP, WebDriver, Live View, or computer-control client. Browserbase keep-alive sessions continue until explicitly released or timed out. Browserless, Kernel standby, and Browserbase keep alive, retrieved 2026-07-26.

What an idle or abandoned session can cost table

Set a timeout below the vendor maximum, close the remote session in a finally block, and log the vendor’s observed time and bandwidth fields. That control matters more than a per-page estimate because a stalled navigation, a disconnected worker with keep-alive enabled, or an agent waiting for input can consume billable time without returning another page. Browserless unit consumption, Browserbase keep alive, and Kernel timeouts, retrieved 2026-07-26.