https://www.youtube.com/watch?v=V-sLwzskvqE

When I was 18, I joined Magic, a frontier lab working on 100M token context coding models, as a reinforcement learning researcher.

While I worked there, I gradually grew more drawn to autonomous research. I truly felt that this could automate all the work I was doing at Magic, so in January, I left to start my own lab.

Introducing Aster — a major breakthrough for research.

Aster isn’t a single agent. It is the first autonomous lab orchestrating thousands of research agents in parallel.

We pointed Aster towards ProteinGym, a benchmark that measures how accurately models predict the effects of protein mutations. It set the world record in 30 minutes.

And it did it efficiently:

    •    57x faster than single agent systems on the same task.

    •    One-third the cost for the same result.

But here’s what we realized: measurable problems are only a small sliver of research.

The hard ones were never measurable.

At Aster, we’re on a mission to automate open-ended research, research that defines its own problems before it can answer them.

The future doesn’t wait to be defined by benchmarks. Neither do we.

Read our research writeups at https://asterlab.ai.