The open benchmark for cloud browsers - and Notte is leading it.
Browser Arena is one harness, one workload, every provider measured the same way. Latency, reliability, and cost across the cloud browser cohort. Built by Notte, runnable by anyone - and on the current run, Notte sits at the top of the leaderboard with 100% reliability, 394 ms latency, and $0.05 per browser-hour.
Top of the leaderboard. Reliability + latency + cost in one number.
100 of 100 sessions. Anchor, Browser Use, and Hyperbrowser each dropped one.
End-to-end session lifecycle. 3x faster than Browserbase, 9x faster than Browser Use.
Tied for the cheapest browser-hour in the cohort. Half the price of Browserbase.
The cloud browser category - Notte, Kernel, Steel, Browserbase, Anchor, Browser Use, Hyperbrowser - keeps publishing latency and reliability numbers that no one else can reproduce. Browser Arena is the answer to that: one harness, one workload, one cohort, every provider measured the same way. Notte built the harness, runs in it like everyone else, and currently leads the leaderboard by Value Score.
The harness measures the four steps that define a usable remote browser: create a session, connect over CDP, navigate to a URL with domcontentloaded, and release the session. The same script runs from the same AWS region against every provider. Results are emitted as JSONL with the per-stage breakdown so anyone can audit them.
We picked this workload because it is what every agent loop pays for. Long browsing sessions amortise create-and-release cost; agentic loops do not. If your agent opens a fresh session per task, the create-and-release floor is your latency floor, and a provider's marketing latency is almost never that number. On the current run, Notte's number is 394 ms - second on raw latency, first once cost is folded in.
Browser Arena is open source on GitHub. Anyone can submit a provider, run their own cohort, or fork the methodology. We publish raw JSONL alongside the aggregates because aggregates lie and tails matter. The Value Score on the leaderboard is a transparent blend of reliability, latency, and cost - the three things that actually decide whether an agent can ship on a given provider.
Notte is one of the providers in the benchmark and, at the time of writing, the one in first place. 100% reliability across 100 runs, 394 ms end-to-end latency, $0.05 per browser-hour, Value Score 0.904. We are explicit about cases where Notte does not lead - cold-start in some regions, specific workloads outside the harness - and we keep the repo open so the numbers can be challenged.
End-to-end session lifecycle (create + connect + goto + release), 100 runs per provider, ranked by Value Score - a blend of reliability, latency, and cost. Notte is first, with Kernel close behind. The next three providers carry latency 3x to 12x higher than Notte at higher or equal cost.
| # | Provider | Region | Reliability | Latency | Cost/hr | Value score |
|---|---|---|---|---|---|---|
| 1 | Notte | us-west-2 | 100.0% | 394 ms | $0.05/hr | 0.904 |
| 2 | Kernel | us-east-1 | 100.0% | 341 ms | $0.06/hr | 0.889 |
| 3 | Steel | us-east-1 | 100.0% | 1,190 ms | $0.10/hr | 0.794 |
| 4 | Browserbase | us-east-1 | 100.0% | 557 ms | $0.12/hr | 0.781 |
| 5 | Anchor Browser | us-east-1 | 99.0% | 3,664 ms | $0.05/hr | 0.761 |
| 6 | Browser Use | us-east-1 | 99.0% | 4,538 ms | $0.06/hr | 0.749 |
| 7 | Hyperbrowser | us-east-1 | 99.0% | 1,761 ms | $0.10/hr | 0.741 |
Snapshot of the public nottelabs/browserarena leaderboard. Live data, including per-provider history and per-stage breakdown, lives on browserarena.ai. Older runs are kept under /history.
Open methodology
One workload, one region per provider, one harness. Per-stage timings published as raw JSONL so every aggregate can be reproduced or contested - including Notte's first-place number.
Reliability-first
Reliability is the gate before latency or cost. Notte and Kernel are the only providers in the current run with no dropped sessions across 100 attempts.
Provider-contributable
Open repo, open PRs. Vendors can add themselves, fix integration bugs, or propose new workloads. Notte built it, but the benchmark is not Notte's to grade.
If you ship agents on cloud browsers, the bottom line is simple: Notte is currently the best price-performance browser infrastructure for AI agents in the open benchmark. Run the harness against your own workload to confirm it. If you operate a cloud browser, submit a PR. If you write about the category, the raw data is yours to cite.