What AI Browsers Cost When Jobs Arrive in Bursts
Compare nine cloud browser pricing models and calculate how peak concurrency and short sessions change an AI agent's monthly bill.

Cloud browsers for AI agents can cost as little as $0.02 per browser-hour, but that rate rarely describes the whole bill. Providers may also charge for traffic, browser creation, completed tasks, or the number of browsers running together. Browser Use lists $0.02 per hour with a one-minute minimum; Cloudflare Browser Run charges Workers Paid users $0.09 per browser-hour after 10 included hours and $2 per additional browser above 10 averaged daily peaks. A workload with the same 300 browser-hours can therefore produce very different browser charges depending on when its jobs run. The examples below are arithmetic from public rates, not invoices or performance tests.
The product boundary matters. A cloud browser hosts a session that Playwright, Puppeteer, or an agent can control. Model calls, search calls, and code execution can add charges beyond browser runtime; check what the selected plan includes. Stagehand, for example, is a framework that can drive a browser; counting it as another browser host would double-count the infrastructure layer.
Compare nine browser billing models
The table records a specific published entry route for each service, read on November 10, 2026. Rates and allowances use different units, so the rows are not a normalized cheapest-to-most-expensive ranking. A limit on simultaneous sessions also says nothing about the number of browser-hours included in a plan.
| Browser service and route | Published browser meter and commitment | Capacity and other meter to check |
|---|---|---|
| Browserbase Developer | $20/month includes 100 browser-hours; then $0.12/hour | 25 concurrent browsers; 1 GB of proxy traffic, then $12/GB. Startup is $99/month with 500 hours and 100 concurrent browsers. |
| Browser Use Cloud | $0 monthly subscription; $0.02/browser-hour, billed by minute with a one-minute minimum | Starts at 10 simultaneous browsers; higher account limits depend on total paid. Home-IP traffic is $5/GB; direct or your own proxy is $0.20/GB. |
| Steel Launch | $0/month plus $0.10/browser-hour; $30 one-time usage credit | 10 concurrent sessions and 15-minute maximum session. Proxy bandwidth is $10/GB; Scale is $250/month plus usage, with 100 sessions and a one-hour maximum. |
| Kernel Developer | $0/month plus usage, with $5 monthly usage credits; rate details list $0.0000166667/second headless and $0.0001333336/second headful | Five concurrent browsers. Headful and headless are materially different charges; Kernel says idle standby time and its managed proxy are not separately charged. |
| Browserless Free | 1,000 units/month; one unit covers up to 30 seconds of an open browser session | Two concurrent browsers and two-minute maximum session. Unit rules add proxy and successful CAPTCHA units. |
| Anchor Starter | $50/month includes 50 credits; $1 per credit overage; its card lists 0.1 credit per completed task | Up to 25 concurrent browsers. The published task charge is not a browser-hour rate, so count tasks and any separately metered resources under the actual account terms. |
| Hyperbrowser Startup | $30/month includes 30,000 credits; browser time is 100 credits, or $0.10, per hour | 25 concurrent sessions; proxy data costs 10,000 credits, or $10, per GB. Agent steps and model tokens are additional if used. |
| Cloudflare Browser Run Sessions | Workers Paid: 10 browser-hours/month, then $0.09/hour | 10 browsers included on the monthly average of daily peaks, then $2/additional browser. Workers Paid has a $5/month minimum and Worker usage may add charges. Quick Actions use a different browser-only meter. |
| Bright Data Browser API | Browser API is traffic-based rather than browser-hour based; the product pricing page lists $8/GB pay as you go | The product page offers 1 GB in its free tier and says concurrent sessions are unlimited. Premium domains can cost more per GB; verify the applicable domain rate before budgeting. |
Browserless’s units are not dollars. Its current page shows Prototyping at $25/month billed annually with 20,000 monthly units and $0.0020/unit overage. That is an annual commitment displayed as a monthly equivalent, not a monthly cancellable $25 plan. An open browser session consumes one unit per 30 seconds, with partial increments rounded up. A residential proxy adds six units per MB; a datacenter proxy adds two. Convert the actual workload to units before comparing its plan with an hourly service.
Kernel’s headless rate works out to roughly $0.06 per hour by multiplying its per-second rate by 3,600. Its headful rate similarly approaches $0.48 per hour. Those are arithmetic equivalents at the published rate, before credits. They do not establish that the two modes are interchangeable for an agent that needs a visible browser, managed authentication, or a particular control method. These hourly equivalents use Kernel’s browser-specific documentation, which labels the rates per second. Its main pricing page instead describes usage in GB-seconds and calls the entry plan Free, while the documentation calls it Developer. Confirm the applicable browser meter in your account before treating these equivalents as a quote.
Record when each session runs
A monthly total of browser-hours loses two facts needed for a quote: how many sessions overlap and how many billing minimums they trigger. Keep a row per session with its creation time, confirmed stop time, provider-reported billed duration, concurrent browser count during the run, proxy traffic, model calls, retries, and whether the intended task actually succeeded. Derive the peak concurrent count for each day from overlapping intervals. Do not use the average number of running browsers as a substitute for the daily peak on Cloudflare.
Check the provider’s usage record after a stop request. The request alone does not prove that a remote browser closed or that time metering ended. On Browserless, the unit rules explicitly charge for time a browser remains open, including idle time. On Kernel, standby mode changes what is charged while idle. A session lifecycle log helps you apply each rule to the right interval.
Record traffic separately. A browser that opens many image-heavy pages can move the bill more than its short runtime suggests on a per-GB service. Bright Data cannot be compared with an hourly route until you have observed GB transferred for the same permitted workload. Likewise, a hosted browser’s bill does not automatically include the reasoning model. Use the AI API cost calculator for the token portion and the AI API price comparison to choose the model rate. Add the result beside, not inside, the browser charge.
The same 300 hours can have a different bill
Consider three illustrative 30-day schedules for Cloudflare Browser Run Sessions on Workers Paid. Each uses exactly 300 browser-hours. Schedule A runs 10 browsers for one hour every day. B runs 100 browsers for six minutes every day. C runs 100 browsers for one hour on three days and no browsers on the other 27 days. These are timing patterns, not measured demand or equivalent delivery deadlines.
| Illustrative schedule | Browser-hours | Monthly average of daily peaks | Browser-hour overage | Concurrency overage | Browser Run components |
|---|---|---|---|---|---|
| A: 10 × 1 hour × 30 days | 300 | 10 | $26.10 | $0 | $26.10 |
| B: 100 × 0.1 hour × 30 days | 300 | 100 | $26.10 | $180 | $206.10 |
| C: 100 × 1 hour × 3 days | 300 | 10 | $26.10 | $0 | $26.10 |
The browser-hour calculation is (300 - 10 included hours) × $0.09 = $26.10. Cloudflare defines concurrency as the monthly average of daily peaks. For B, (100 × 30) ÷ 30 = 100, so (100 - 10) × $2 = $180. For C, (100 × 3 + 0 × 27) ÷ 30 = 10, so the additional concurrency component is zero. Cloudflare totals daily browser seconds and rounds the monthly sum to a whole hour; 300 exact hours do not require an extra rounding assumption.
These figures exclude the required $5 Workers Paid minimum, any Worker request or CPU overage, and any other products used. More importantly, C finishes jobs on only three days. It is useful only if the real queue can wait, and neither a cheaper schedule nor a higher permitted peak proves a service will meet a deadline. Browserbase and Steel instead publish concurrent-session limits by plan; a workload above such a limit needs a larger plan, queueing, or a different architecture. A cap is different from Cloudflare’s averaged concurrency charge.
Cloudflare’s limits also matter to this illustration: Workers Paid defaults to 200 concurrent browsers and three new instances per second. A 100-browser burst must ramp up under that launch limit, and a browser closes after 60 seconds of inactivity unless its timeout is extended or it stays active. The schedules assume the sessions can be kept open for their stated durations; they do not promise instantaneous starts or task completion.
Short sessions change the effective hourly rate
Now hold actual running time at one hour on Browser Use. One 60-minute session uses one browser-hour, so its browser-time component is $0.02. Split the same running time into 360 isolated 10-second sessions. Each crosses the published one-minute minimum, so the billed duration is at least 360 × 1 minute = 360 minutes = 6 browser-hours. Its browser-time component is at least 6 × $0.02 = $0.12, before traffic, model calls or any credit balance. Browser Use publishes the minute billing and minimum; this example does not assume how it rounds a 61-second session.
This sixfold difference applies to the browser component in this constructed case. It does not justify combining unrelated users’ sessions: login state, privacy, isolation and task deadlines may require separate browsers. Nor does a $0.12 component mean the account owes $0.12 in cash if prepaid credits remain. A credit top-up is a payment into a balance; consumption is a separate event.
Browserless rounds each browser session’s open time up to a 30-second unit. Splitting the same runtime across separate short sessions can therefore consume more units than one continuous session. This rule does not establish a new charge for reconnecting to an existing session. Its documentation also says failed work may consume the browser time and proxy traffic used before failure. Do not apply Browser Use’s one-minute minimum to Browserless or assume all failed runs are free.
Check eligibility before comparing totals
Build the shortlist from peak demand and longest session first. Browserbase Developer allows 25 concurrent browsers and 100 included hours at $20/month; Startup allows 100 and 500 at $99/month. A 100-browser burst cannot be justified by the Developer plan’s low monthly fee even if monthly hours stay below 100. Steel Launch’s 10-session and 15-minute limits similarly rule out some schedules before its $0.10 hourly rate matters. Hyperbrowser Startup and Anchor Starter each publish 25 concurrent sessions; their included credits use different units.
Browser Use lists 10 simultaneous browsers at the starting account level and unlocks 50 after $100 total paid, 250 after $1,000, and more at higher lifetime payment thresholds. Those are eligibility thresholds on the account, not monthly concurrency charges. Confirm the account’s actual unlocked limit and available credit balance before promising a burst capacity. If a provider publishes no rounding or overage rule for a particular mode, leave that cell unknown rather than treating a blank as free.
For a complete budget, price the accepted task rather than merely an opened browser: browser time or units, bandwidth, plan commitment, task or creation fees, model tokens, and retries. The web search API comparison covers search calls that may precede navigation; AI sandbox pricing covers code workspaces that may persist after a browser stops. Neither is a second browser bill. Test a small representative workload and read the provider’s usage export before relying on an estimate; no vendor success-rate or latency comparison was run for this article.
Sources
- Browserbase pricing: plans, included hours, concurrency and proxy traffic.
- Browser Use pricing: browser time, minimum, traffic and paid unlocks.
- Steel pricing: Launch and Scale rates, credits and limits.
- Kernel pricing and rate details: plans, per-second modes and standby.
- Browserless pricing and unit consumption: plan units, rounding and extra usage.
- Anchor pricing: credits, completed tasks and concurrent browsers.
- Hyperbrowser pricing: browser credits, traffic and capacity.
- Cloudflare Browser Run pricing and Workers pricing: browser hours, daily-peak average and plan minimum.
- Cloudflare Browser Run limits: default concurrency, launch rate and idle timeout.
- Bright Data Browser API and maker pricing explanation: traffic-based billing, published rate and premium-domain exception.
All pages read on November 10, 2026.
Free toolWhat will 10,000 chats a month cost?Estimate my bill

