Hyperbrowser vs. Steel: managed scrape-scale or open-source agent runtime?

Hyperbrowser and Steel both pitch themselves as "infrastructure for AI agents," and the surface-level feature lists overlap heavily — cloud Chromium over CDP, stealth, proxies, CAPTCHA, session recordings. The substantive differences live in two places: Hyperbrowser is closed-managed and bets on parallelism plus TLS-layer fingerprint stealth; Steel is open-source-first with a self-hostable runtime and bets on agent-shaped state primitives (Profiles, Credentials, Files, Agent Logs). Public benchmarks now put Steel ahead on standing and reliability, which moves the decision squarely onto deployment shape and runtime primitives.

At a glance

What is Hyperbrowser?

Hyperbrowser is a serverless cloud-browser platform that runs fleets of stealth-patched Chromium instances in isolated containers, accessible via WebSocket/CDP as a drop-in replacement for local Playwright, Puppeteer, or Selenium. Two architectural bets define it. The first is concurrency: 1,000+ concurrent baseline, 10,000+ "instant," burst from 0 to 5,000 in under 30 seconds, and 50,000+ for flash-sale events. The second is TLS-layer stealth: Stealth Mode and Ultra Stealth Mode ship JA3/JA4 TLS fingerprint randomization to bypass Cloudflare/Akamai, plus navigator.webdriver patching and mouse-curve randomization for behavioral resistance.

The product surface includes HyperAgent (a thin AI layer with page.ai(), page.extract(), executeTask()), a native MCP server, residential proxies in 170 countries, dedicated US/EU static IPs, and BYOIP for enterprise IP-block control. Pricing is credit-based with per-second billing, and a browser-hour lands mid-pack among the measured providers. Startup is $30/mo for 30,000 credits and 25 concurrent. SOC 2 Type II + HIPAA, 12 global regions, 99.99% uptime claim, 180-day enterprise retention. The ICP is engineering teams running scrape-scale workloads or parallel AI agents.

What is Steel?

Steel is an open-source browser API positioned for AI agents — "Humans use Chrome. Agents use Steel." It ships a managed cloud and an open-source steel-browser runtime that customers can self-host (Docker, Railway 1-click, bare-metal Node.js, build-from-source) with the same API as Steel Cloud. The opinionated design choices map to AI-agent needs: Profiles for persistent identity (cookies, extensions, localStorage, fingerprints, up to 30 days, 300 MB cap), a Credentials API with AES-256-GCM per-record + KMS re-encryption (TOTP, blur, autoSubmit, exactOrigin), a Files API for artifacts, MP4/HLS replay (replaced rrweb), live view via WebRTC at 25fps capturing OS-level dialogs, mobile mode with real touch and viewport, and Agent Logs that tie tool calls to the replay timeline.

Pricing is tiered by concurrency, browser-hours, and retention rather than per-second metered: Free $10 credits ≈ 100 browser-hours; Start $29/mo (290 hrs, 10 concurrent); Developers $99/mo (1,238 hrs, 20 concurrent); Pro/Startups $499/mo (9,980 hrs, 100 concurrent). Steel Cloud is SOC 2-compliant; Steel Local is the self-host option. The runtime is agent-framework-neutral — it works with Playwright, Puppeteer, Selenium, Browser Use, Stagehand, and any agent that speaks CDP. Native integrations include Hermes (Nous Research), Pi/OpenClaw, and Browser Use.

How they compare

Speed and reliability: both mid-pack, both reliable

Per the public Browser Arena leaderboard (browserarena.ai), Steel ranks #3 while Hyperbrowser ranks #7 of seven measured browser runtimes. Steel is faster and fully reliable in the current public run; Hyperbrowser trails on both raw speed and value score, though its hourly cost is still mid-pack rather than Browserbase-level expensive. Browser Arena is maintained by Notte Labs but the benchmark is open-source and reproducible on Railway across current public run. Steel's own historical browserbench harness reports 0.89s avg / 1.09s p95; that figure is stale by the current independent run. Hyperbrowser's "sub-second launch" pitch is similarly stale.

The honest read for either pick: speed isn't the wedge in this pair. Steel is clearly ahead on the public leaderboard, but the decision rests elsewhere — on OSS self-host, identity primitives, and concurrency ceiling. If your workload is "1,000 short-lived browsers in 30 seconds," Hyperbrowser's burst architecture is the wedge regardless of single-session latency. If your workload is "10 long-lived agents that need a runtime built around their state," Steel's primitives are the wedge.

OSS self-host vs. closed managed

This is the sharpest single-axis differentiator. Steel's steel-browser open-source runtime is the strongest wedge — Docker, Railway 1-click, bare-metal, build-from-source, and the same API as Steel Cloud. Steel Local has a practical limitation worth knowing: it's effectively single-session, no managed stealth, no Credentials/Files API, no managed proxies — those are Cloud-tier features. But for teams with a hard data-residency or auditability requirement, the OSS option is the differentiator that no Hyperbrowser offering can match.

Hyperbrowser is closed-managed-only. The trade-off Hyperbrowser makes in return is operational scale — 10K-50K concurrent, multi-region, BYOIP, JA3/JA4 stealth at the network layer, 99.99% uptime claim — capabilities that are non-trivial to replicate in self-host. Picking Steel for the OSS option means trading away that operational ceiling unless you're operating at the Steel Cloud Pro/Startups tier (100 concurrent).

Identity, state, and stealth shape

Both ship stealth defaults; the shapes differ. Hyperbrowser's stealth bets are at the network and TLS layers — JA3/JA4 randomization, residential proxies, BYOIP, dedicated US/EU static IPs. Steel's stealth bets are at the runtime and state layers — Profiles preserve fingerprint coordination across sessions, Credentials API handles auth without exposing secrets to the agent, mobile mode ships a real fingerprint and touch-event model rather than a UA string. Steel's anti-bot and CAPTCHA APIs exist but the proxy supply at lower tiers is less network-layer-aggressive than Hyperbrowser's.

The practical read: if your bot-detection problem is at the TLS handshake, Hyperbrowser. If your problem is "this agent needs to log in and stay logged in across sessions while not leaking secrets to the LLM," Steel. The Skyvern review of Hyperbrowser flags that auth flows (2FA/TOTP) require developer build on Hyperbrowser; Steel's Credentials API supports TOTP natively.

Compliance and observability

Both are SOC 2 Type II. Hyperbrowser adds HIPAA. Steel ships MP4/HLS replay (replaced rrweb), Agent Logs as a first-class object tying tool calls to replay timestamps, and live view at 25fps via WebRTC that captures OS-level dialogs and PDF viewers. Hyperbrowser ships session recordings, console log streaming, Playwright Trace Viewer integration, live remote attachment, and Datadog/New Relic integration. Both support 180-day-class retention at enterprise tiers (Hyperbrowser 180-day, Steel 14-day at Pro/Startups, custom at Enterprise).

Pricing models

Hyperbrowser's credit-based per-second model meters browser uptime and proxy bandwidth, with browser-hours landing mid-pack and proxy traffic metered separately. Steel's tiered pricing meters concurrent sessions and browser-hours, with the effective per-hour cost compressing as you scale up tiers. Hyperbrowser's price is flatter per browser-hour but adds metering complexity at the credit layer. Steel's framing of itself in head-to-heads is "predictable monthly spend and straightforward capacity planning"; Hyperbrowser's is "no per-GB billing surprises like BrightData." Both are accurate descriptions of their own model — the right pick depends on whether your finance team prefers tier ceilings or per-second metering.

When to choose Hyperbrowser

• You need 10K-50K concurrent sessions for scrape-scale or flash-sale-style workloads.
• TLS-layer fingerprint randomization (JA3/JA4) is the wedge against your bot-detection problem.
• BYOIP, static IPs, and 170-country residential proxies are operational requirements.
• Your team wants closed-managed and is happy with credit-based per-second billing once usage stabilizes.
• HIPAA compliance + SOC 2 Type II + 180-day enterprise retention map to your regulated workload.

When to choose Steel

• You need an OSS option — Docker self-host, full data residency, audit-the-runtime.
• Agent-shaped state primitives (Profiles, Credentials, Files, Agent Logs) are core to your workflow rather than nice-to-haves.
• You want 24-hour sessions with explicit sessions.release() discipline rather than concurrency-burst architecture.
• Token efficiency matters and Steel's mobile mode + Markdown output APIs (claimed up to 80% LLM token reduction) move the cost needle.
• You're integrating Hermes, Pi/OpenClaw, or Browser Use natively and want a peer-level browser provider.

A third option: Notte

A third option worth a mention here is Notte (notte.cc), a cloud Chromium platform purpose-built for AI agents. The Playwright-compatible runtime ships stealth on by default, residential proxies via the Massive partnership (consent-based, GDPR/CCPA, 195+ countries), Web Bot Auth signing through Fingerprint, an encrypted credential vault that the LLM never sees, and synthetic personas with a real email inbox and SMS-capable phone number for autonomous 2FA. Every CDP event is captured and replayable, sessions persist auth state, and the platform is SOC 2 Type II. Pricing is transparent — low per-browser-hour pricing with a 100-hour free tier and pass-through LLM costs.

Notte ties Hyperbrowser on SOC 2 Type II compliance and ranks #1 overall on the public Browser Arena leaderboard — a tier ahead of Steel at #3 and Hyperbrowser at #7. The wedge for Notte is value (top-tier speed at the lowest hourly cost) plus identity-in-the-runtime: Vault and Personas (with real inbox + SMS for autonomous 2FA) are productized primitives Hyperbrowser leaves to network-layer stealth and Steel ships at the Credentials API level only. Web Bot Auth signing via Fingerprint's Bot Directory is unique to Notte at this fidelity, and the Massive proxy partnership is consent-sourced (Hyperbrowser ships residential proxies; Steel's lower tiers BYO).

Verdict

Hyperbrowser and Steel both market themselves as infrastructure for AI agents, but the design choices diverge cleanly. Hyperbrowser bets on closed-managed scale, TLS-layer stealth, and operational ceilings (10K-50K concurrent, BYOIP, multi-region). Steel bets on OSS self-host parity, agent-shaped state primitives, and predictable tiered pricing. Per the public Browser Arena leaderboard, Steel sits at #3 while Hyperbrowser sits at #7; Steel is faster and more reliable in the current public run, but the real wedge is still product shape rather than a single speed reading.

Pick Hyperbrowser when concurrency-at-scale or TLS-fingerprint stealth is the wedge. Pick Steel when OSS self-host or agent-shaped runtime primitives are the wedge. If you want the #1 Browser Arena value score, the lowest hourly cost among measured providers, productized identity primitives (Vault, Personas, Web Bot Auth), and consent-sourced Massive proxies in one runtime, Notte is the third option to evaluate.