Browser automation pricing has become genuinely hard to compare in 2026. As AI agents take over web tasks — clicking, filling forms, and scraping sites that have no API — a wave of providers has emerged, and no two charge the same way.
Some bill per token. Others bill per credit, per “unit,” per gigabyte, or per browser-hour. They also sit at very different points on the open-source-to-fully-managed spectrum.
This guide breaks down browser automation pricing and features across the major players so you can compare them side by side.
Why Browser Automation Pricing Is So Hard to Compare
Browser automation pricing is hard to compare because there are two kinds of tool in this market, and conflating them is the first mistake buyers make.
The first kind is a hosted browser — the actual cloud infrastructure that runs a real Chromium instance for you. The second is an orchestration layer — an open-source CLI or framework that drives someone else’s browser.
Disparity in the Browser automation pricing units make things worse — A token, a credit, and a “unit” all mean different things, and the cheapest option depends entirely on the shape of your workload. Short bursty tasks favor unit-based plans; long sessions favor per-hour billing; proxy-heavy scraping is dominated by per-gigabyte bandwidth cost.
Browser Automation Pricing and Features: Full Comparison Table
The table below covers all 14 providers across pricing, open-source status, concurrency, stealth, and the AI agent layer.
| Provider | Type | Pricing | Open Source | Concurrency | Proxies | CAPTCHA | Stealth / Anti-bot | AI / Agent Layer | Protocol / Compatibility | Self-Hostable | Notable Extras | Maturity |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Rusty Browser | Hosted browser | Free (175 tokens/mo); $10/$50/$150 per mo; PAYG $0.05/hr | Yes — MIT | Unlimited | Premium, unlimited use included | Included, all users | “Ghost Stealth” adaptive evasion | Natural-language AI instruct engine | WebDriver BiDi / CDP, direct Chromium | Yes (cluster via Flux) | Results storage, JS hydration, custom parsers, auto-scaling | Very early (v0.1.0) |
| Browser Use | Hosted browser + agent | $0/$29/$299/$999 per mo; $0.06/hr | Yes — open library | 3 → 500 by tier | $5/GB | Included | Advanced stealth | “Browser Use 2.0” custom model + v3 agent | Built on Playwright | Partial (harness) | ~15× cheaper, 6× faster than frontier models; BYO-key | Established (1.3M+ downloads) |
| BrowserBase Stagehand | Hosted browser + framework | Free/$20/$99 per mo; ~$0.10–$0.12/hr | Yes — MIT (framework) | 3 → 250+ by tier | $10–$12/GB | Auto, paid tiers | Basic → Advanced by tier | act/extract/observe/agent primitives | Direct CDP; TS + Python | Stagehand runs local | Action caching, self-healing, session replay, Model Gateway | Established |
| Browserless | Hosted browser | Free (1k units); $25/$140/$350 per mo + Enterprise | Core closed; clients open | ~1 → 100 by tier | Residential, 6 units/MB | 10 units/solve | BrowserQL anti-fingerprinting | Smart Scrape API, MCP server; no AI agent | Puppeteer/Playwright via WebSocket | Yes (Enterprise) | Live debugger, session replay, Lighthouse testing, hybrid automations | Mature |
| Steel.dev | Hosted browser | Free ($10 credits, 100 hrs); $29/mo, $99/mo | Yes — fully open source | Thousands of sessions | Built-in proxy chain / IP rotation | Built-in | Anti-flagging controls | Foundation only — you build agent logic | Puppeteer, Playwright, Selenium, CDP | Yes — deploy or local | Page→markdown/PDF, session viewer, extensions; 24-hr sessions | Public beta |
| Skyvern | Hosted browser + agent | Free open-source; cloud usage-based (credits) | Yes — open source | Parallel instances (cloud) | Bundled in cloud | Bundled in cloud | Anti-bot in cloud version | Vision-LLM + planner-actor-validator loop | Playwright-compatible SDK | Yes (pip / Docker) | No selectors — computer vision; no-code workflow builder; templates | Established |
| Hyperbrowser | Hosted browser | Free tier; credit-based (~$0.10/hr); ~$30/mo for 30k credits | HyperAgent framework open source | 10,000+ concurrent browsers | Global IP rotation | Automatic | Fingerprint randomization (UA, canvas, WebGL) | HyperAgent (Playwright AI extension) | Puppeteer, Playwright, Selenium | No (managed) | Sub-ms latency, 99.9% uptime, isolated containers, sync/async clients | Newer, fast-growing |
| Kernel | Hosted browser | Free tier; usage-based, no idle charges, proxies never billed | Yes — open source | Fleet-based (org limits) | Residential, never billed separately | Automatic | Built-in stealth mode | Agent runtime / code execution platform | Playwright, Puppeteer, Browser Use, Magnitude | Yes (open infra) | Unikernel, sub-150ms cold starts, live view + replay, persistent auth, SOC 2 / HIPAA-ready | Newer (YC S25, $22M raised) |
| Anchor Browser | Hosted browser | Monthly plan + usage: per-browser, per-hr, proxy/GB, per AI step | No | Unlimited concurrent browsers | VPN integration, proxy support | Automated CAPTCHA resolution | Custom session fingerprinting, anti-bot bypass | “Perform web task” natural-language endpoint | CDP, Playwright, APIs, agent frameworks | No (cloud-hosted) | Full browser isolation, Okta/Azure AD, SSO/RBAC/DPA, MCP-first; batch up to 5,000 | Newer |
| Bright Data Agent Browser | Hosted browser | Free (2 GB), Starter $39/mo (10 GB), PAYG $8/GB; enterprise $500–1,000+ min | No | 1M+ concurrent sessions | Huge residential/datacenter/mobile network | Autonomous CAPTCHA solving | Headful GUI browser harder to detect; advanced unlocking | /agent endpoint takes plain-English tasks | Puppeteer, Selenium, Playwright | No | ~98% success across hard domains; MCP server; web data layer | Mature (enterprise) |
| Notte | Hosted browser + framework | Usage-based (single API) | Open framework (SSPL license) | On-demand headless instances | Standard + customizable proxies | Bundled | Browser interface for LLM agents | Full-stack web AI agents; perception layer | CDP, cookie integration | Yes (open framework) | Secure credentials vault, session replay, websites→navigable maps | Newer |
| Vercel Agent Browser | Orchestration CLI | Free (open source); pay only the underlying provider you connect | Yes — open source (vercel-labs) | Depends on connected provider | Depends on connected provider | Depends on connected provider | AI chat via Vercel AI Gateway | Drives Chrome, Lightpanda, AgentCore, Browserbase, Browserless, Browser Use, Kernel | Yes (local CLI) | Snapshot refs (@e1/@e2) for LLMs, auth vault, domain allowlist, live dashboard | Newer (Vercel Labs) | |
| Lightpanda | Browser engine | Free; open source | Yes — open source | Lightweight local engine | Bring your own | Bring your own | Minimal footprint | None built-in | CDP-compatible | Yes (local engine) | Headless browser built for speed/low memory; used as an engine, not a service | Newer |
| AWS Bedrock AgentCore Browser | Hosted browser | AWS usage-based (pay per use) | No | AWS-scale | Via AWS networking | Not bundled | Standard | Part of Bedrock AgentCore agent runtime | CDP / Playwright | No (AWS-managed) | Enterprise AWS integration, IAM, VPC, secure sandboxes | Newer |
| Scrapybara | Hosted desktop/browser | Usage-based (credits) | No | Parallel instances | Bundled | Bundled | Standard | Computer-use sandbox for agents | Computer Use APIs (Anthropic/OpenAI) | No (managed) | Full virtual desktop, not just a browser; built for computer-use agents | Newer |
How to Read the Browser Automation Pricing Above
Treat the pricing column as a starting point, not a verdict.
Vercel’s Agent Browser shows “$0” because it’s an orchestration layer — it doesn’t run browsers itself, so you still pay whichever hosted provider you connect it to. Lightpanda is free for the same reason: it’s an engine, not a service.
Among the genuinely hosted browsers, the cheapest headline rate is rarely the cheapest total bill. Proxy fees, CAPTCHA charges, and idle-time billing can quietly double a quote.
You can verify any figure directly on each vendor’s own pricing page — for example Browserbase pricing, Browser Use pricing, and Browserless pricing. Underlying control protocols are documented in the Chrome DevTools Protocol reference and the W3C WebDriver BiDi spec.
Bottom Line: Which Provider to Pick
When it comes to browser automation pricing, there’s no single winner — the right pick depends on your workload.
For a full autonomous agent that plans and reasons, start with Browser Use or Stagehand. For messy legacy or canvas-heavy sites where the DOM is unreliable, Skyvern‘s vision approach earns its higher per-step cost.
For maximum control and zero lock-in, the open-source pair Steel.dev and Kernel lead — Kernel on latency, Steel on maturity. For deterministic scraping, QA, and document rendering, Browserless is a dependable veteran.
For enterprise scale, Bright Data (deepest anti-detection) or AWS AgentCore (if you’re already on AWS) are the conservative picks. Scrapybara is the odd one out — a full virtual desktop for computer-use agents, not just a browser.
Two warnings on browser automation pricing: figures change fast, so confirm them on the provider’s own page before committing. And don’t trust the marketing tables — including each vendor’s own — until you’ve run a small proof-of-concept against your real target sites.
Comments
No comments yet. Be the first to share your thoughts!
Leave a Comment