browser-use-olympics
Browser-agent benchmark with five timed events from one prompt, plus a ~200-line DevTools agent where Jev picks each action.
Overview
This repository holds two things. Browser Use Olympics is a benchmark for browser-using agents hosted on Almond: every agent gets the same prompt, visits a page that carries five events, and is timed by a server-side clock between Start run and Finish run, with a Hall of Fame. Each run draws its own identities from a seed but keeps the same shape.
Almond-fastloop is a dependency-free, roughly 200-line browser computer-use loop and the reference entry. Each tick it reads the page through Chrome DevTools (interactive elements with names, values and a diff, in about 2 ms), builds the list of executable options, and asks Jev one choice plus two yes/no questions before an executor acts. The README reports fastloop completing 5 of 5 events in about 13 to 14 seconds with a Claude planner before the clock, and gives a cost breakdown at list prices with Jev's share at about $0.0018.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which executable option on the page should be taken next? | — |
| Yes / no | Two yes/no checks per tick (e.g. is the event done?) | yes · no |
How it works
- Planner reads the task before the clock
- Each tick, DevTools snapshots interactive elements
- Build executable options and ask Jev one choice and two yes/no questions
- Executor acts; repeat until finished
Strengths
- Small, dependency-free agent
- Public benchmark with server-side timing
- Transparent cost reporting
Caveats
- Benchmark hosted on a commercial platform
- Reference entry is by the benchmark authors
Getting started
Store TYPESAFE_API_KEY in ~/.config/typesafe/env and run almond-fastloop.mjs / olympics.mjs.At a glance
- Type
- Project
- Status
- beta
- Language
- JavaScript
- Runtime
- Node + Chrome DevTools
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Browser Use Olympics by Almond: one prompt, five events, one clock. Plus fast loop, a ~200-line browser computer-use agent (Chrome DevTools + TypeSafe Jev).
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Which option to take each tick, plus two yes/no checks.
How fast was the reference agent?
About 13-14 seconds for 5 of 5 events, per the README.
Can any agent enter?
Yes, all agents get the same prompt.
Related projects
ego-decision-layer: measured, pluggable decision layer for ego lite where Jev picks operation and target in one request.
ProjectbetaTypeScriptPlugs any CDP browser, including engines that never draw a page, into a Jev loop that picks operations and targets from the DOM.
LibrarybetaPythonBrowser automation where Jev picks every step from a sightmap's named actions and reports whether the goal is met.
ProjectbetaGoBrowser agent that acts on one look at the page with one open-model call per action; Jev is an optional chooser backend.
AppprototypePythonOpen Apache-2.0 browser-agent decision model (Qwen3.5-35B-A3B fine-tune) that doubled Jev 1.13's task success in the same agent.
ServiceresearchPythonBrowser driver where Jev makes every decision and agent-browser performs every action, benchmarked against Claude Code on 10 tasks.
BenchmarkbetaTypeScript