For Computer Use, the harness matters as much as the model

MiguelG719

Hi HN, Over the last 2 years, we observed computer use models improving at a rapid pace and saturating benchmarks. This new benchmark replaces Online-Mind2Web with our own Browserbase Benchmark v2 that better represents the complex tasks that browser agents face in the real world. It runs against 23 models (frontier and open-weight) and 9 harnesses (Claude Code to LangChain Deep Agents) on accuracy, speed, and cost. This new benchmark confirmed our belief that the choice of an harness is becoming as important as the choice of a model. For example: claude-opus-5 runs 74% at $1.50/task on LangChain deep agents but 71% at ~$10/task on fx. The eval harness is a CLI you can run yourself (pick harness + tools/mcps + model, pass high-level tasks, grades with LLM verifiers, has trials/concurrency/OTEL tracing): https://github.com/browserbase/stagehand/tree/main/packages/... Happy to get into methodology, and if you want your model or harness added, just let me know.

peeet

I love the website, it would be 10/10 if I could go to chrome://dino

mhykim

Why use Stagehand when agents can write CDP / Playwright on the fly for browser use

starlightttt

Love the design

smpandya

Are there results comparing agents running different tools (agent-browser, playwright MCP, browse CLI), or is this mostly Stagehand focused?

pranaygup12

What model family do you find is the best for browser use overall? or does it change pretty regularly

dericdinudaniel

Cost difference across harnesses is interesting to me. Would love to see more info about more optimizations in the harnesses to trim down costs.

devk03

Hyper personalized harnesses are the edge that the labs cannot beat startups on. There will be a whole entire era of new harnesses coming out soon.

jay_sahnan

Why invest in browser agents if computer use like Astra is already so good at solving tasks?

sharath39

Nicely built.