Skip to main content

​
Per-task, not per-minute

Most browser-automation APIs bill by the minute of sandbox time. That pushes the cost of a 10-second extraction to the same price as a 10-minute wait — and the vendor has no incentive to make anything faster. Browser Agents bills per task:
  • Each task costs at least 0.1 credits (= $0.01)
  • Tasks that run longer bill at 10× our raw compute cost
  • Whichever is greater wins
For typical extraction tasks (~20 seconds), the floor dominates and you pay $0.01/task. For long multi-step flows (~1 minute), compute scales linearly.

​
Credit math

  • 1 credit = $0.10
  • Floor per task = 0.1 credits = $0.01
  • Margin over raw compute = 10×
  • credits_billed = max(0.1, raw_cost_usd × 10 / 0.10)

​
Raw cost composition (for the curious)

On an average 22-second Daytona task:
Line itemCost
Daytona compute (22s × ~$0.067/hr)$0.00041
Gemini LLM (~2 planner calls)$0.00030
Mongo storage + ops$0.00034
S3 (screenshot PUTs + 30d storage)$0.00004
Total raw~$0.0011
Billed: $0.01 (floor). Effective margin: ~9×.

​
Comparison

Measured $/passing-task on the full 101-task benchmark (2026-04-15T15-40-08):
ProviderBilling model$ / passing task
Browser Agentsper-task, $0.01 floor$0.010
Browser Use Cloudper-request$0.126
Firecrawl Interact7 credits/min + 1 credit/scrape$0.0092
Firecrawl Scrape (no actions)1 credit/scrape$0.0057
Browser Use Cloud is 12.6× more expensive per passing task than Browser Agents. Firecrawl Interact is slightly cheaper per-task but its pass rate is 16 tasks lower and it hits per-account concurrency caps at real load.

​
Volume

  • First 100 tasks: free with any signup
  • Solo + enterprise pricing: see pricing page
  • Enterprise volume discounts available

​
What you pay for

  • Running the sandbox (Daytona/E2B/Modal — cheapest-first fallback)
  • The planner LLM’s input + output tokens
  • Screenshot storage in S3 (auto-expires)
  • Our infrastructure (queue, fallback routing, etc.)

​
What you don’t pay for

  • Failed cold starts we retry
  • Sandbox delete time (we fire-and-forget)
  • Scheduler overhead

​
Provider cost transparency

Under the hood, Browser Agents runs across three providers. Current rates per vCPU·hour:
Provider$/hr (1 vCPU + 1 GiB)
Daytona$0.067
E2B (1 vCPU)$0.067
Modal$0.166
We route to the cheapest available and fall back on infrastructure failure. You pay the same $0.01/task floor regardless of which provider actually ran your task.