Stagehand just crossed 1 Million weekly downloads, but we're not done yet.
New in our latest 3.5.0 release:
- native clipboard API
- screenshots in extract()
- better snapshots & local mode use
We’re graduating from Playwright.
We’re grateful it gave devs the foundation to build fast and adopt even faster, but fit changed as we scaled.
Stagehand’s next chapter goes deeper into the browser stack.
We did a complete rewrite of Stagehand to run directly on CDP.
By decoupling from Playwright, we’re now 44% faster.
Work with any automation framework, efficiently traverse iframes & shadow DOMs, and even auto-cache agentic workflows.
Stagehand Agent can now use MCP tools.⚒️
Simply pass in the server url, and tell the agent to use the tools.
Watch Stagehand use Exa, Supabase, Notion, and Stripe MCPs. 🧵
Today @GoogleDeepMind released a state of the art computer use model, in partnership with Browserbase.
Computer use is hard to evaluate. You need reliable browser infrastructure and realistic tasks.
Below, we cover how we ran these benchmarks and how you can try yourself!
Last week we launched Stagehand v3, built directly on CDP.
We fully decoupled from Playwright, and became 44% faster.
Here's one of the engineers @mcguiresean_ to explain where the speed gains are coming from:
If you haven't heard... Stagehand is now officially a python library! 🐍
Bring the best browser automation library into your tech stack.
pip install stagehand 🤘
The new GPT-5 performs worse than Opus 4.1 in Stagehand evals in both speed and accuracy.
The smaller models are faster, but also still fall short of Opus 4.1.
We've built the best in house observability for Stagehand 🔭
Get inference time, token usage, LLM output, and more on the new dashboard.
No more black box, see everything behind the scenes.
See for yourself, get started with npx create-browser-app 🤘
🚨 Stagehand Evals Leaderboard Update🏆
OpenAI's open source models on @GroqInc are very fast and accurate.
With an average inference cost of $0.003 per task and 86% overall accuracy on our benchmark, it's the most performant model from @OpenAI
