Stagehand (@Stagehanddev) — posts, photos & videos

@Stagehanddev

Claude 5.5 Haiku is on par with GPT 6 Luna from OpenAI on accuracy and speed, but is 3x the cost per task.

As computer use improves, most gains will show up on the Pareto frontier: maintaining high accuracy while also being faster and cheaper.

2 2 0 10 485 4

@Stagehanddev reposted

@whereischarly

0 1 2 10 1.1K 5

@Stagehanddev

Stagehand supports @ClaudeDevs' new Tool SDK, converting model actions into browser actions.

Follow our quickstart guide to build your first browser agent with Claude and Stagehand.

github.com/browserbase/cl…

ClaudeDevs@ClaudeDevs · Oct 7

Computer use and browser use toolsets are now built into the Python and TypeScript SDKs for Claude.

The API tells you what Claude wants to click or type.

Previously, you had to write your own loop and map clicks and keystrokes to commands, but the SDKs now run the loop and send actions to drivers.

1:00

Open quoted post →

3 3 3 21 4.5K 9

Stagehand

@Stagehanddev

Oct 1

Computer use models are getting really fast.

Opus 5 (released in July '26) averages 678s per task while newer models like Opus 5.5 average 326s, ~2x speed up.

See the speed up for yourself across 72 model x harness combinations at stagehand.dev/evals

4 4 0 20 1.4K 4

Stagehand

@Stagehanddev

Sep 23

GPT-6 Luna is a step function improvement from it's predecessor scoring 11% higher in accuracy on our browser agent tasks.

GPT-6 Sol scores 2% better in accuracy, but costs 20x more per task than Luna.

OpenAI Developers@OpenAIDevs · Sep 22

GPT-6 Sol and Luna just landed in Astra’s orbit.

Both launch today with API prices 50% lower than GPT-5.6.

Build with Sol. Scale with Luna. To production and beyond.

0:15

Open quoted post →

6 2 1 17 1.6K 2

Stagehand

@Stagehanddev

Sep 22

Stagehand can use Stripe's new WebMCP tools in checkout pages.

Run page.tools() to see what WebMCP connectors a page exposes. With Stripe, you can use Stagehand to add items to cart and complete the checkout with native tools.

0:13

Jeff Weinstein@jeff_weinstein · Sep 22

Hi internet! We went ahead and upgraded all @stripe hosted checkout pages for 7.8M businesses—0.45% of the world’s GDP—with WebMCP so agents can more efficiently buy. Our evals show this reduces checkout latency by 39% and token usage by 42%.

Learn more: stripe.dev/blog/how-strip…

Open quoted post →

3 4 2 28 6.9K 12

Stagehand

@Stagehanddev

Sep 22

During early testing on our Browser Agent Evals, Opus 5.5 completely crushed it's predecessor Opus 5, but also Fable 5.1 & 5 in accuracy, speed, and, cost.

Compared to Opus 5 in the Claude Code harness it's roughly 5x cheaper per task, and 2x faster.

Claude@claudeai · Sep 22

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.

It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.

0:20

Open quoted post →

8 2 0 18 1.4K 5

Stagehand

@Stagehanddev

Sep 18

Everyone can build Agentic Commerce using the @link and Browse CLI powered by Stagehand!

Paul Klein IV@pk_iv · Sep 18

it seems like agentic commerce is finally solved with stripe link. and we're excited to partner with stripe to enable all agents to transact online.

your agent can now natively make payments on the web using link cli + browserbase.

one shot prompt: docs.browserbase.com/integrations/s…

0:26

Open quoted post →

0 3 0 20 3K 5

Stagehand

@Stagehanddev

Sep 15

Muse Spark 1.3 from @Meta tops the charts in our Browser Agent benchmarks using the @mastra agent harness.

In this harness, Muse outperforms Opus 5 in cost, speed, and accuracy.

19 7 2 79 8.1K 28

@Stagehanddev

0 2 3 33 6.8K 27

@Stagehanddev

Introducing our Browser Agent Evals.

We benchmark frontier and open models using various harnesses on computer-use tasks.

Our evals are fully open source and reproducible via our Evals CLI.

0:08

10 10 3 38 7.1K 16

Stagehand

@Stagehanddev

Sep 1

We got early access to Fable 5.1 and evaluated it extensively across our internal benchmarks.

It outperformed all other frontier models (including Fable 5) on accuracy scoring 92.11%, while being 60% cheaper than Fable 5 per task.

Claude@claudeai · Sep 1

We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.

They're the world’s most advanced models for coding and knowledge work.

0:27

Open quoted post →

5 5 2 25 2.8K 3

Stagehand

@Stagehanddev

Aug 27

FYI: Stagehand supports WebMCP as a first-class primitive. just call 𝚙𝚊𝚐𝚎.𝚝𝚘𝚘𝚕𝚜() in your agent!

happy hacking :)

0:26

Chrome for Developers@ChromiumDev · Aug 25

Create agent-ready web apps for the WebMCP Challenge → goo.gle/4qpDQYV

We're excited to see ChatGPT supporting WebMCP, the experimental standard that lets users and AI agents navigate websites together. In the @OpenAIDevs WebMCP Challenge you can experiment with it and win prizes. Plus, Chrome's @sarah_edo is on the judging panel! What will you create?

Open quoted post →

2 11 1 87 12.6K 67

Stagehand

@Stagehanddev

Aug 21

Stagehand v4 gives you more control over how your browser actions are cached.

With Browserbase caching, your scripts automatically run up to 80% faster.

Stagehand@Stagehanddev · Aug 10

Introducing Stagehand v4: the SDK for browser agents.

Playwright was built for testing, we built Stagehand for your agent: with improved context management, self-healing actions, and iframe support.

1:29

Open quoted post →

2 4 2 35 4K 8

Stagehand

@Stagehanddev

Aug 17

Last week we launched Stagehand v4, the fastest and most token efficient way to control a browser.

We benchmarked it extensively and found that Stagehand repeatedly outperforms Playwright in every action and in every region on Browserbase.

Stagehand@Stagehanddev · Aug 10

Introducing Stagehand v4: the SDK for browser agents.

Playwright was built for testing, we built Stagehand for your agent: with improved context management, self-healing actions, and iframe support.

1:29

4 1 0 17 1.5K 3

@Stagehanddev

2 4 4 50 7.5K 36

@Stagehanddev

Introducing Stagehand v4: the SDK for browser agents.

Playwright was built for testing, we built Stagehand for your agent: with improved context management, self-healing actions, and iframe support.

1:29

53 45 25 426 267K 389

Stagehand

@Stagehanddev

Jul 28

Introducing our Model Router, get the most accurate model for the task at the cheapest price.

Customers using the router are seeing up to 30% lower token costs with little to no change in overall accuracy.

0:06

4 5 2 29 6K 9

Stagehand

@Stagehanddev

Jul 14

Stagehand 3.7 is out.

Here's what shipped last week:
- Control where your agent can navigate to domain allowlisting/denylisting
- CUA support for Gemini 3.5 Flash and the GPT-5.6 family (Sol, Luna, Terra)
- Various bug fixes & patches

pnpm add @browserbasehq/stagehand@latest

1 2 0 13 1.3K 5

Stagehand

@Stagehanddev

Jun 24

Gemini 3.5 Flash + CUA is impressive: on our benchmarks, it performs similarly to frontier models while being on average ~80% cheaper.

It's a reliable model for powering browser agents that balance speed and accuracy on real-world tasks.

Google AI Studio@GoogleAIStudio · Jun 24 Open quoted post →

2 6 0 16 3.2K 4