trajectory playback

Live Lab

Watch how frontier AI agents solve real optimization tasks. Every decision, every benchmark, every pivot — replayed from actual runs. See where agents struggle, adapt, and break through.

161
runs
23
tasks
7
models
663M
tokens
> watch_a_run> propose_task
// contribute

Propose a Task

Can models begin to participate in the loops that drive scientific and engineering progress? Every task you propose helps answer that question. Submit a task with a clear metric, an open search space, and no known optimal solution — and watch how frontier models run the empirical improvement loop.

Every trajectory in Live Lab started with someone proposing a task. Three tiers to match your time and resources.

industrial benchmark

High Tier

Full benchmark-ready package. Directly adopted into the sandbox.

skeleton frame

Mid Tier

Instruction + base code. We build the rest.

concept phase

Low Tier

Just describe the pain point. No code needed.

> propose_a_task