Playwright vs Stagehand (2026): What ChatGPT, Claude & Gemini Say

AI models name Playwright more often. In 50 recorded answers, Playwright was named in 48 and Stagehand in 41. Playwright was named first in 21 answers. Stagehand was named first in 7. The named counts are close. The named-first counts are not.

This page counts what AI answers name for browser automation for agents. It does not judge either product.

TL;DR

• Expect both names in almost any AI answer on this topic. The real gap is which one comes first, and that is mostly Playwright.
• Claude Fable 5 is the only model that names Stagehand more often than Playwright. The Anthropic family names Stagehand in all 15 of its answers (Stagehand 100%).
• Gemini 3.5 Flash is the sharpest split: Playwright in 5 of 5 answers, Stagehand in 2 of 5. If your team asks Gemini 3.5 Flash or Perplexity, Stagehand may not come up.
• Choose by the job, not the count. Smoketest’s comparison assigns known, stable paths to Playwright and volatile UI to Stagehand’s act().
• Budget from your own runs. Neither vendor record in this edition holds pricing.

How often do AI models recommend Playwright and Stagehand?

Playwright leads on every measured line. The gap shows most in first position, not in presence.

Answer share is the share of the 50 recorded answers that named the product. Named first is the count of answers where it appeared before any other tracked product. Playwright is also the category leader, so no reference row is needed. Stagehand shares its 41 named answers with Browserbase, which was named first once.

The two products sit close on presence and far apart on placement. Stagehand’s named-versus-first gap is 68 points. Playwright’s is 54 points. So the models include Stagehand in most answers but usually put something else ahead of it. When either one is named, it lands near the top of the answer: average position 2.58 for Playwright and 2.9 for Stagehand. The full category record is on the browser automation for agents index.

Which models prefer Playwright, and which prefer Stagehand?

The OpenAI models, Gemini 3.5 Flash and Perplexity lean Playwright. Claude Fable 5 leans Stagehand. The other models name both equally often, and the Anthropic family as a whole names Stagehand in every answer.

OpenAI. GPT-5.6 Sol, ChatGPT and GPT-5.6 Luna each name Playwright in every answer and Stagehand in 4 of 5. At family level that is Playwright (100% of OpenAI answers) against Stagehand (80% of OpenAI answers). The lean is on presence only. Several OpenAI answers open with Stagehand as the headline pick, as the quotes below show.

Anthropic. This is the one family where Stagehand leads. It is named in all 15 Anthropic answers (Stagehand 100%), against Playwright (93.3% of Anthropic answers). Claude Fable 5 drives the difference. It names Stagehand in 5 of 5 answers and Playwright in 4 of 5, and Stagehand is its most-named product (Stagehand 100%). Claude Opus 5 and Claude name both in every answer.

Google. The two Gemini models disagree sharply. Gemini names both in every answer. Gemini 3.5 Flash names Playwright in 5 of 5 and Stagehand in 2 of 5, the weakest Stagehand result on the panel. Family level: Playwright (100% of Google answers), Stagehand (70% of Google answers).

Perplexity. Perplexity names Playwright in 5 of 5 and Stagehand in 3 of 5. Sonar Reasoning Pro names each in 4 of 5, and its most-named product is neither of them. It is Browser Use (100% of Sonar Reasoning Pro answers). Family level: Playwright (90% of Perplexity answers), Stagehand (70% of Perplexity answers).

The sharpest disagreement is Claude Fable 5 against Gemini 3.5 Flash. The first puts Stagehand ahead of Playwright. The second names Playwright in every answer and Stagehand in 2 of 5.

What do the answers say about each?

The quoted answers place the two at different layers of one stack more often than they set them against each other.

“Use Playwright as your foundation.” ChatGPT

“For most production AI agents, I’d choose Stagehand, ideally backed by Playwright and a managed browser service when necessary.” GPT-5.6 Sol

“Playwright is the deterministic base most agent frameworks build on.” Claude Opus 5

“use Browser Use if you’re in Python or Stagehand if you’re in TypeScript” Sonar Reasoning Pro

“Traditional browser automation tools like Playwright and Selenium were built for deterministic end-to-end testing” Gemini 3.5 Flash

In the ChatGPT, GPT-5.6 Sol and Claude Opus 5 answers, Playwright is the control layer and Stagehand, where named, is the AI layer on top. Sonar Reasoning Pro sets Stagehand against Browser Use, not Playwright, and splits them by language. The Gemini 3.5 Flash answer frames Playwright as a testing tool. That model also names Stagehand least often.

How do Playwright and Stagehand differ?

They differ in cost structure and in the job each is named for. Neither vendor record in this edition holds pricing, so the pricing detail below comes from the captured pages.

Pricing model

No pricing is recorded for either product in this edition’s vendor records.

Playwright. A Playwright selector step carries no inference cost. No captured page states a licence or price for Playwright itself.

Stagehand. Stagehand is an open-source library from Browserbase. An uncached Stagehand AI action costs an LLM inference. A cached rerun replays recorded actions without touching a model. Run on Browserbase, browser time is sold as monthly plans with included browser hours. Model inference there is billed pay-as-you-go through Browserbase’s Model Gateway.

The practical difference: Playwright selector steps add no model spend. Stagehand adds a metered model call wherever an action is not cached.

Who each is for

Playwright. Microsoft released Playwright. It works across Chrome, Firefox and Safari. Its steps are deterministic. Browserbase argues that Playwright’s auto-waiting is built for testing, not automation. That view comes from the company that makes Stagehand.

Stagehand. Stagehand’s AI actions keep working through UI changes that break selectors. It exposes four primitives: act(), observe(), extract() and agent(). The current version talks to the browser directly over the Chrome DevTools Protocol, not through Playwright. Playwright stays compatible and can drive the same browser session.

What the models name each for

In the recorded answers, Playwright is named for deterministic browser control. ChatGPT’s stack table assigns reliable, deterministic browser control to Playwright. Stagehand is named for natural-language actions and structured extraction. GPT-5.6 Luna recommends Playwright plus Stagehand and credits Stagehand with the AI-native actions.

When should you pick Playwright?

Pick Playwright when your agent follows paths you can script, or when you want the name the models place first.

• Known, stable paths. Smoketest’s decision table assigns a known, stable path to Playwright selectors.
• Cross-browser coverage. Playwright runs on Firefox and Safari as well as Chrome.
• No per-step model spend. Selector steps carry no inference cost.
• Broad model coverage. Named in every answer by all but two models, and those two name it in 4 of 5.
• First placement. Named first in 21 of 50 answers, the most of any product in the category.

Where Playwright leads: presence across every family and first placement.

When should you pick Stagehand?

Pick Stagehand when the pages your agent visits change often, or when you want natural-language actions without giving up coded steps.

• Volatile UI. Smoketest assigns redesigns, A/B tests and third-party widgets to act().
• Mixed control. Stagehand can take over a Playwright-driven page for the messy middle of a flow.
• Repeatable AI steps. Stagehand can cache AI actions on Browserbase’s server or in a local directory.
• Browserbase users. Browserbase describes Stagehand as integrating with Browserbase.
• Claude-heavy research. Every Anthropic answer names it, and Claude Fable 5 names it more often than Playwright.

Where Stagehand leads: the Anthropic family, and Claude Fable 5 in particular.

How this sits against the Playwright vs Stagehand guides

The captured pages compare architecture and features. This page counts names in AI answers. Each captured page has an authorship or commercial interest, named below.

Smoketest, “Stagehand vs Playwright, and When You Need Both”. It argues for using both, with Playwright on known paths and act() on volatile spots. It says descriptions of Stagehand as a wrapper around Playwright are out of date. Smoketest sells a monitoring product that runs this same hybrid. It declines to repeat per-action dollar figures it could not trace to a primary source.

Hacker News thread. The captured item is a comment from a user writing as Stagehand’s builders. It announces a fix for round-trip latency, which it calls Stagehand’s biggest flaw. One reply describes friction between Playwright’s web-testing scope and task automation.

Browserbase, “Choosing between Playwright, Puppeteer, or Selenium?”. This page is vendor-authored. Browserbase makes Stagehand. It ranks Playwright above Puppeteer and Selenium. It calls Stagehand the AI-powered successor to Playwright. It describes Stagehand’s AI APIs as sitting on top of the base Playwright Page class. Smoketest reports that Stagehand has since moved off Playwright onto the Chrome DevTools Protocol.

Two more ranked results, a Reddit thread and a LinkedIn post, could not be fetched and are not used here.

The distinction: those pages tell you how the tools work. This page tells you which one ten AI models name, how often, and which model leans which way. None of the captured pages carries a per-model split.

How the sample was built

10 models x 5 fixed prompts = 50 recorded answers. Edition 2026-09. The panel tracks 17 vendors in this category and 15 were named.

The five questions, verbatim:

1. “What is the best browser automation tool for AI agents? Name specific products.”
2. “Which browser automation tool would you recommend to AI agents in 2026?”
3. “Compare the top browser automation tool options right now.”
4. “I’m AI agents and I need a browser automation tool. What should I use and why?”
5. “Best browser automation tool for AI agents to let an agent use websites?”

Families and answers: OpenAI (GPT-5.6 Sol, ChatGPT, GPT-5.6 Luna) gave 15. Anthropic (Claude Opus 5, Claude, Claude Fable 5) gave 15. Google (Gemini, Gemini 3.5 Flash) gave 10. Perplexity (Perplexity, Sonar Reasoning Pro) gave 10. The labels ChatGPT, Claude, Gemini and Perplexity refer to GPT-5.6 Terra, Claude Sonnet 5, Gemini 3.6 Flash and Sonar Pro in the run record. The full method is on the method page, and every recorded answer is on the category index.

What these counts cannot tell you

The counts measure presence in AI answers. They say nothing about which product is better, faster, cheaper or more reliable. Being named is not the same as being recommended, because an answer can name a product only to warn against it. Each model answered each prompt once. This is one dated snapshot, edition 2026-09. Names are matched as strings, so a passing mention counts the same as a headline pick. API answers can differ from what the consumer chat apps show. The prompts are in English. All five ask about AI agents, so a buyer choosing a tool for end-to-end testing is outside this sample.

Frequently asked questions

Is Playwright going to replace Selenium?

The panel does not forecast. In this category, AI models named Playwright in 48 of 50 answers and Selenium in 24 (Selenium 48%). Browserbase’s guide ranks Playwright above Selenium. Browserbase says it supports all three frameworks it compares. Read its ranking as a vendor view, not a measurement.

What are the disadvantages of using Playwright?

The captured criticism comes from Stagehand’s maker. Browserbase argues that Playwright’s auto-waiting adds latency when a browser is driven for automation rather than testing. It also says Playwright lacks stable frame identifiers. The panel measures neither claim.

Which is faster, Playwright or Puppeteer?

None of the captured pages benchmarks the two. The panel counts names, not speed. In this category, Puppeteer was named in 24 of 50 answers (Puppeteer 48%) and never named first. Browserbase’s guide says Puppeteer needs more manual waiting than Playwright. That is a point about code, not speed.

Is Playwright really better than Selenium?

The panel does not measure better. It measures naming: Playwright was named first in 21 of 50 answers, Selenium in 1. Browserbase’s guide calls Playwright the best of the three frameworks it compares. Browserbase makes Stagehand, so read that as a vendor view.

Can you use Playwright and Stagehand together?

Yes. Playwright can connect to Stagehand’s browser over the Chrome DevTools Protocol. Several recorded answers name the pair as one stack, with Playwright for control and Stagehand for AI actions.

Is Stagehand still built on Playwright?

No, according to Smoketest. The current Stagehand talks to the browser over the Chrome DevTools Protocol, with playwright-core only an optional peer dependency. Browserbase’s own earlier guide describes Stagehand’s AI APIs as sitting on top of the Playwright Page class.

Does Browserbase make Stagehand?

Yes. Browserbase describes Stagehand as its open-source library. The models name the two equally often: Browserbase in 41 of 50 answers (Browserbase 82%), the same as Stagehand. Browserbase was named first once.