Skip to main content

workload / Checked 2026-11-02

Choosing a browser platform for bursty browser workers

Many jobs arrive at once and need browsers, while most other times have little activity. Feature and limit support for all 12 providers, linked to their documentation.

About this workload

Published by Notte, which competes with the products on this page. Every figure links to the provider's own documentation, checked on 2026-11-02.

Many jobs arrive at once and need browsers, while most other times have little activity.

Browser slots and session creation rate constrain the burst independently. Queue work before allocating a browser, and apply bounded retries to avoid a synchronized retry wave.

Acceptance test: Measure time to drain a fixed burst, total attempted starts, task success and the bill for both successful and failed work.

Provider support

Statuses and limits for the features this workload depends on, as stated in each provider's documentation.

ProviderMax concurrencySession creation rate limitBilling unit
NotteBrowsers: 5 Free, 25 Developer, 100 Startup, Custom Enterprise. Agents and functions: 5, 25, 100, Custom.[1]Rate limited per API key with a separate budget for session starts (POST /sessions/start): per minute 50 (Free), 50 (Developer), 100 (Startup); per hour 200, 200, 1,000; per day 600, 600, 10,000; Enterprise custom. Requests over the limit return 429.[2]Dollar-denominated credits spent per hour of browser session ($0.01 per hour); the page says usage is rounded by a billing policy without stating the increment[1]
BrowserbasePer plan: Free 3, Developer 25, Startup 100, Scale 250+. Assigned at the organization level and split across projects.[3]Per plan, per any 60-second period: Free 5, Developer 25, Startup 50, Scale 150+. Requests over the limit receive a 429 error.[3]Browser time is billed by the minute (one-minute minimum per session); proxy bandwidth by the MB (one-MB minimum per session). Plans include a monthly allocation of browser hours and proxy GB; overage is billed per browser hour and per GB.[4]
KernelOrg-wide: Developer 5, Hobbyist 10, Start-Up 150, Enterprise custom; reserved browser pool capacity and browsers in standby count toward the limit[5]A per-organization browser create rate limit set by plan; the numeric rate is not published. Exceeding it returns 429 with Retry-After and the SDKs retry up to 2 times. Acquiring from a browser pool is not subject to it. Enterprise limits are custom[5]Per second of active browser runtime. A browser that is idle for 5 seconds after the last CDP or Live View activity enters Standby Mode and stops accruing usage cost; usage rates are published in USD per second[6]
Steel10 on Launch, 100 on Scale, 1,000+ on Enterprise[7]Not publishedper browser hour, billed by the minute and rounded up[7]
Hyperbrowser1 on Free, 25 on Startup, 100 on Scale, 1,000+ on Enterprise[8]Not publishedcredits: 100 credits per hour of browser usage ($0.10 per hour), plus proxy data at 10,000 credits per GB ($10 per GB). 1 credit = $0.001.[9]
Anchor BrowserUp to 5 on Free, 25 on Starter, 200 on Growth, 500+ on Enterprise (Team: 50 per the marketing page). Hosted MCP docs separately say there is no concurrency limit.[10]Per minute: Free 5, Starter 25, Team 50, Growth 100, Enterprise 200+. POST /v1/sessions and POST /v1/sessions/async each have their own budget; 429 with Retry-After is the rate limiter, 429 without it is the concurrency cap. Recording fetch is capped at 2,000 requests per team per hour.[10]Credits ($1.00 per credit overage). Browser sessions you drive yourself: $0.05 per browser hour on active time rounded to the nearest full minute, plus $0.01 per browser created; Anchor proxy $8 per GB; AI steps $0.01 to $0.10 each. Automation Task runs: flat 0.1 credits ($0.10) per run.[11]
Browser Use Cloud10 concurrent sessions on a new project; 50 / 250 / 500 / 1,000 after $100 / $1,000 / $5,000 / $25,000 total paid; 1,000+ on Enterprise. Agents and browsers share one limit.[12]No separate session-creation limit is documented. HTTP request budget per project: creates, lists, updates, deletes, event reads and full run reads get max(25, 2 x stored concurrency) requests/second, capped at 100; selected status reads get max(25, 2 x stored concurrency) with no cap. A Tier 0 project is on the 25 RPS floor. The project limiter counts requests in five-second windows. Separate per-IP edge ceilings apply: 1,000 requests/second for general API traffic and 2,500 requests/second for selected status reads, evaluated over 300 seconds.[13]Browser time: $0.02 per browser-hour, metered by the minute. Credits for the requested timeout are reserved when the browser is created and unused reserved credits are returned after it ends. The final charge has a one-minute minimum and partial minutes round up. Network traffic is charged separately per GB, and hosted agent runs also bill model tokens.[12]
BrowserlessPer plan: Free 2; Prototyping 10 annual / 5 monthly; Starter 40 / 30; Scale 100 / 80; Enterprise custom[14]Not publishedUnits: 1 unit per 30 seconds of open browser session, partial 30 s increments rounded up. Proxy traffic (per MB), successful CAPTCHA solves and some APIs also draw from the same unit balance.[15]
Cloudflare Browser RunWorkers Free 3 concurrent browsers; Workers Paid 200 per account by default, with higher limits available on request.[16]Browser Sessions: 1 new browser every 20 seconds on Free, 3 per second on Paid. Quick Actions: 1 request every 10 seconds on Free, 30 per second on Paid. Excess requests get HTTP 429 with a Retry-After header.[16]Browser hours for all methods: usage is totaled in seconds per day and the monthly total is rounded to the nearest whole hour. Browser Sessions (Puppeteer, Playwright, CDP) are also billed for concurrent browsers, calculated as the monthly average of the daily peak.[17]
Amazon Bedrock AgentCore Browser1000 concurrent active sessions per account by default, adjustable via support ticket[18]StartBrowserSession: 30 transactions per second per account, adjustable. CreateBrowser: 5 TPS.[18]Per second of consumed CPU (priced per vCPU-hour) plus peak memory (priced per GB-hour), 1-second minimum, 128 MB minimum memory; session spans microVM boot to termination. I/O wait and idle time are free if no background process runs.[19]
LightpandaExplorer 5, Builder 30, Enterprise custom; sessions over the limit are rejected with HTTP 429[20]Not publishedPer browser hour of open session time, metered by the second and summed over the billing period; concurrent sessions are a separate limit[20]
TilionPer plan, counted across the whole account: Free 10, Developer 25, Team 250, Enterprise 250 and up[21]A per-API-key creates-per-minute throttle (rate_limits.session_creates_per_minute in GET /v1/usage); an example Team response shows 60. Per-plan values are not published in the pages read.[21]Browser time is metered for each minute a session is alive (on credit-funded accounts 1 credit per minute, 60 credits per browser hour). The launch post states $0.03 per browser-hour. Developer and Team are card-billed against monthly browser-hour allowances with metered overage.[22]
Sources (22)

notte.cc [1] ↗docs.notte.cc [2] ↗docs.browserbase.com [3] ↗docs.browserbase.com [4] ↗kernel.sh [5] ↗kernel.sh [6] ↗docs.steel.dev [7] ↗hyperbrowser.ai [8] ↗hyperbrowser.ai [9] ↗docs.anchorbrowser.io [10] ↗docs.anchorbrowser.io [11] ↗browser-use.com [12] ↗docs.browser-use.com [13] ↗docs.browserless.io [14] ↗docs.browserless.io [15] ↗developers.cloudflare.com [16] ↗developers.cloudflare.com [17] ↗docs.aws.amazon.com [18] ↗aws.amazon.com [19] ↗lightpanda.io [20] ↗docs.tilion.dev [21] ↗docs.tilion.dev [22] ↗

Keep comparing

Put the workflow to the test.

Try Notte Read the quickstart ↗