Browser Use vs Browserbase: Pay by the Minute or Pay by the Month
Browser Use side by side with Browserbase, in 3 minutes.
Also known as: browser-use, Browser Use Cloud
Open source browser agent library plus a cloud with hosted agents and stealth browsers for AI agents, priced by usage.
Browser Use makes software that lets AI agents operate a real web browser. It began as an open source Python library, MIT licensed, in which a developer describes a task in plain language and the agent navigates, clicks, fills forms and extracts data on sites it has not seen before. The library is free, runs on the developer's own machine against a local or remote browser, supports many model providers including local models, and can be extended with custom tools, lifecycle hooks and tracing. The Browser Use CLI gives Claude Code, Codex or another coding agent control of the user's own Chrome, with its tabs and logins, or of a cloud browser.
Browser Use Cloud sells two products on top. Browser Use Agents (API V4) takes a task and returns the finished result and files, running the agent and its browser in Browser Use's cloud. V4 writes and runs code to finish long, multi step tasks across many pages, and the faster V3 agent acts on the page step by step. Runs chain as follow ups in a session, and rerunnable scripts let a first run save tested code that later runs reuse and repair when a site changes. Browser Infrastructure rents managed stealth Chromium browsers, with CAPTCHA handling, residential proxies, saved profiles, live view and recordings, to any agent or to Playwright, Puppeteer, Selenium or CDP code.
Developers get a REST API with an OpenAPI file, Python and TypeScript SDKs, a CLI and an MCP server. Each run can use a model from OpenAI, Anthropic, Google, xAI, MiniMax and other providers, with GPT-6 Luna as the default, or the customer's own OpenAI, Anthropic or Google key. The cloud takes scheduled work and has an iMessage channel, and a person can step in at a human checkpoint through the live view.
An optional LLM judge returns a verdict on each run, and customers choose a US or EU (Frankfurt) deployment. Enterprise terms add SOC 2 Type II, HIPAA support, a DPA, zero data retention and custom SLAs. There are no roles, SSO or SCIM, and Browser Use describes no memory layer beyond session history and workspace files.
From our comparison videos
Browser Use side by side with Browserbase, in 3 minutes.
Browser Use side by side with Skyvern, in 3 minutes.
Recorded in October 2026. The profile here is live and updates as research is verified.
https://browser-use.com/
Browser / computer-use agent
independent
The open source library takes custom tools, and the Cloud API authorizes connected integrations for agent runs. An MCP server serves Claude Code and other clients, and tutorials cover n8n, Vercel, Claude Managed Agents, OpenClaw and Hermes Agent. Hosted browsers accept Playwright, Puppeteer, Selenium and CDP, and x402 supports pay per request.
Your scraping and automation scripts break every time a target site changes its layout. Browser Use lets your agent reason from the live page instead of hardcoded selectors, so the automation self-heals across redesigns.
You want to automate a site that has no API and sits behind a normal web UI. Browser Use takes a plain-English task and navigates, clicks, fills forms, and extracts data against the real page, model of your choice.
You want Claude Code or Cursor to actually operate websites, not just describe steps. Browser Use ships an MCP server that exposes navigation, form-filling, data extraction, and tab management as native tools to your coding agent.
11.5 / 14 capabilities · 82%
| Integrations & Tool Calling | Full |
|---|---|
|
In the open source library, developers can add custom tools to the agent, and the Cloud API lists integrations by category and authorizes and disconnects connected accounts for agent runs. Sourcedocs.browser-use.com/open-source/customize/tools/addread 2026-09-27 |
|
| Workflow Orchestration | Full |
|
The V4 agent writes and runs code to finish long, multi step tasks across many pages. Cloud runs chain as follow ups in one session with a message queue, and rerunnable scripts let a first run save tested code that later runs reuse deterministically and repair when the site changes. The open source library supports parallel agents, follow up tasks and Playwright steps mixed with agent steps. Sourcedocs.browser-use.com/cloud/agent/scriptsread 2026-10-22 |
|
| Knowledge Grounding & RAG | Not documented |
|
There is no retrieval index over the customer's knowledge. Workspaces hold files a run can read, and the hosted web search tool searches the web at run time. Sourcedocs.browser-use.com/cloud/agent/workspacesread 2026-09-27 |
|
| Human Oversight & Guardrails | Full |
|
A run can include a human checkpoint for approvals, authentication, payments or review. The run stops, a person acts in the live view, and a follow up run continues in the same session. The checkpoint is requested in the task, not enforced by policy. Sourcedocs.browser-use.com/cloud/agent/human-in-the-loopread 2026-09-27 |
|
| Security, Identity & Governance | Partial |
|
Browser Use is SOC 2 Type II audited by an independent firm, with the report under NDA, and offers a DPA with EU SCCs, HIPAA support and zero data retention on enterprise plans. Secrets reach the agent as placeholders, a 1Password vault can fill logins and 2FA codes, and a session can be held to allowed domains. Access is by project API keys and OAuth with a single mcp scope, and there are no roles, SSO or SCIM. Sourcebrowser-use.com/enterpriseread 2026-10-22 |
|
| Observability & Auditability | Full |
|
Each run exposes a paged event stream and status through the API, browser sessions carry live view and recordings for debugging and audit, network events can be captured, and the open source library supports tracing through Laminar and OpenLIT, plus cost tracking. Sourcedocs.browser-use.com/cloud/agent/observabilityread 2026-09-27 |
|
| Memory & State Persistence | Partial |
|
A V4 session holds conversation history across follow up runs, and a workspace keeps files, including scripts the agent saved, across runs until they are deleted. Browser Use does not describe a memory layer with a set scope and lifetime, and saved browser profiles hold browser state, not memory. Sourcedocs.browser-use.com/cloud/agent/workspacesread 2026-09-27 |
|
| Deployment & Data Residency | Full |
|
Customers choose the US deployment or the EU deployment in Frankfurt, each with its own API base URL and separate storage for runs, sessions, files and profiles, and the open source library self hosts on the customer's own machines. Sourcedocs.browser-use.com/cloud/eu-deploymentread 2026-09-27 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
Example apps in the open source docs cover separate jobs such as an ad generator, automated QA, a news monitor and a WhatsApp sender, alongside code templates and agent skills that the customer clones and runs. The Cloud offers no packaged agents ready to adopt. Sourcedocs.browser-use.com/llms.txtread 2026-09-27 |
|
| Triggers & Channel Coverage | Full |
|
Cloud agents take scheduled work as well as API, SDK, MCP and n8n requests, and an iMessage group can be linked to an agent (US beta). Schedules and webhooks are US features not offered in the EU deployment. Sourcedocs.browser-use.com/cloud/which-productread 2026-10-22 |
|
| Model Flexibility & Routing | Full |
|
Each Cloud run takes a model ID from a published list spanning OpenAI, Anthropic, Google, xAI, Z.ai, DeepSeek, Xiaomi, Moonshot AI and MiniMax models, with GPT-6 Luna as the V4 default. The customer's own OpenAI, Anthropic or Google key is an option, and the open source library supports a list of providers that includes local models. Sourcedocs.browser-use.com/cloud/agent/modelsread 2026-10-22 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Developers get a REST API covering V4 runs, sessions, workspaces, profiles, browsers and integrations, with an OpenAPI 3.1 file, along with Python and TypeScript SDKs, a CLI and a hosted MCP server with OAuth. Sourcedocs.browser-use.com/cloud/api-v4-overviewread 2026-09-27 |
|
| Testing, Debugging & Optimization | Full |
|
A V4 run can request a judge, an LLM that evaluates the completed run against criteria the customer supplies and returns a verdict with the run, and sessions carry a feedback endpoint. There is no dataset or regression harness. Sourcedocs.browser-use.com/cloud/faqread 2026-10-22 |
|
| Browser & Computer Use | Full |
|
Browser Use agents drive a real Chromium browser, either Browser Use's hosted stealth browsers with CAPTCHA handling, proxies, profiles and live view, or a local or remote browser through the open source library. The Browser Use CLI lets Claude Code, Codex or another coding agent drive the user's own Chrome with its logins, and hosted browsers also accept Playwright, Puppeteer, Selenium or CDP. Sourcedocs.browser-use.com/cloud/which-productread 2026-10-22 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
BU Ultrafast is open to every Browser Use Cloud plan in the US, including the free plan, and its waitlist is closed. Developers switch to it by setting the model to bu-ultrafast on a run, with no other model settings to tune. A companion option runs the same agent with more reasoning per step.
Bears on: Browser/computer use
View sourceBrowser Use updated its Browser Harness v4 API to support client-controlled session recording, allowing API callers to explicitly enable or disable recording via the browserSettings.record parameter. The release also added a new v3:browsers:read OAuth scope that grants external applications read access to existing browser sessions.
Bears on: Observability / auditability
View sourceBrowser Use released the Browser Harness agent API to the public. The update also introduced new AI models across the platform and lowered the pay-as-you-go top-up minimum.
Bears on: MCP / tool calling / API
View sourceUsage based, at $0.02 per browser hour plus traffic and model tokens. New accounts get a one time $1 credit, and the open source library is free.
Usage based, from a single credit balance in dollars that covers browser time, proxy traffic and model tokens. There are no subscriptions.
Included quota
Free accounts get a one time $1 credit, 10 concurrent sessions and cost efficient models only. Pay as you go bills $0.02 per browser hour, metered by the minute, plus $5 per GB for the managed residential proxy or $0.20 per GB for direct or own proxy egress, and model tokens at each model's published rate (GPT-6 Luna, the V4 default, is $0.12 in and $0.60 out per million tokens). A customer's own OpenAI, Anthropic or Google key carries a 0.2x orchestration fee instead. Concurrency tiers open at $100, $1,000, $5,000 and $25,000 of lifetime spend, for 50, 250, 500 and 1,000 sessions. Enterprise is a negotiated annual credit pool with 1,000+ sessions, custom SLAs, HIPAA support and zero data retention.
What is public
Browser Use publishes machine readable pricing. There are no subscriptions, and new accounts get a one time $1 signup credit. Pay as you go is $0.02 per browser hour, $5 per GB for the managed proxy or $0.20 per GB for egress, and per model token rates. Concurrency is set by lifetime spend, and Enterprise is a negotiated annual credit pool. The open source library is free.
Billing mechanics
A single credit balance in dollars covers browser time, traffic and model tokens. Browser time is reserved for the requested timeout and unused credit is returned, with a one minute minimum. With the customer's own model key, the provider bills tokens and Browser Use adds a 0.2x orchestration fee. Legacy subscriptions can still be managed from the billing page.
Cost watchouts
Model tokens can outweigh browser time at $0.02 an hour, especially on premium models. Managed proxy traffic costs $5 per GB. The 0.2x orchestration fee applies even with the customer's own model key.
Variable cost rationale
Three meters draw on one credit balance at once. Browser time is $0.02 an hour, billed by the minute, proxy traffic is billed by the GB, and model tokens bill at each model's rate. Spend grows with concurrency, run length and token volume, and model choice moves the token line most. Browser Use Agents bill model tokens at 1.2x their published rate, while a customer's own key pays the provider directly plus a 0.2x orchestration fee. No subscription or commitment sets a floor. Credits are prepaid top ups that never expire, automatic recharge is optional, and concurrency is capped by lifetime spend, from 10 sessions up to 1,000. Enterprise moves to a negotiated annual credit pool.
Additional watchouts
Three meters (browser time, traffic and model tokens) share one balance, so model choice drives most of the cost. Free accounts are limited to cost efficient models until a first top up, and concurrency depends on lifetime spend.
Overage / add-ons
There is no plan allowance to exceed. Credits are bought as top ups from $5 to $100,000, or in presets of $100, $500 and $1,000, never expire and can recharge automatically. Concurrency rises with lifetime spend, from 10 to 1,000 sessions.
Sales call required
Mixed (some tiers require a call)
Free / trial
One time $1 credit for eligible Google, GitHub or Microsoft signups with no card required, and the open source library is free
Lowest paid plan
Pay as you go credit top up from $5
Commercial notes
Browser Use sells two cloud products, Browser Use Agents (API V4) and Browser Infrastructure, beside its open source library, CLI and MCP server. Enterprise terms add SOC 2 Type II under NDA, a DPA with EU SCCs, HIPAA support, zero data retention and EU residency on request.
Key ambiguities
Pricing is component based rather than a plan list, so no single monthly figure describes it. Legacy subscription tiers are no longer sold.
Cancellation / refund
No subscription or commitment, and top up credits never expire. Enterprise is a negotiated annual agreement.
Support SLA / resale
Docs and community support on pay as you go. Enterprise adds dedicated support and custom SLAs.
Missing data
Enterprise pool pricing is negotiated, and everything else is published.
The closest documented capability profiles to Browser Use among browser and computer-use agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded
Found a vendor we missed? Have feedback on the index? We'd love to hear from you.