//RESEARCH / SEPTEMBER AI

Researching and building powerful AI.

We research and build powerful AI, closing the open problems between what it can do today and what it will take to get there. The work is to build the capability and prove it where people feel it.

Explore the research Join the team

Building Powerful AI

//RESEARCH PILLARS

The capability, and the good it does.

We research and build powerful AI, working through the open problems it takes to get there, then point the capability at where the result reaches the most people: lower costs, real discoveries, capability that does not stop at the door of a top lab.

Pillar 01

Capability That Does the Work

Powerful AI

We research and build powerful AI: capability that does real work in the world, not just answers in a chat. Reaching it means solving open problems no one has closed yet, and those problems are the work we take on.

We do the research and the engineering that turn raw capability into work you can stand behind, on a path where the safe option and the capable option are the same.

Capability you can put to work
Safe and capable, one path
Judged on results, not promises

Pillar 02

The Work That Compounds

Outcomes That Reach People

Capability only matters when people feel it. We point powerful AI at the bottlenecks that ration real progress: the hours in a day, the experiments a team can run, the cost of a treatment or a year of school. The work is to move those, and to make sure the gains reach the people usually last in line, not only the few.

We measure what the work actually produces, feed it back into the research, and publish what holds up.

Lower costs where people feel them
Reaches everyone, not just the few
Gains that last and stack

//RESEARCH PRINCIPLES

Capability is racing ahead. Powerful AI is not here yet.

Capability Is Racing Ahead. Powerful AI Is Not Here Yet.

Raw ability is climbing fast, but the open problems between it and powerful AI are still unsolved, and closing them is the work that remains. We measure ourselves by what arrives and how fast. Let others chase benchmarks; we chase outcomes we can show you.

We Move Fast, And We Are Honest About The Limits.

We push to close the gap between what AI can do and what it could do for people. The physical world, the gaps in data, and the patience of institutions set the real pace, and we plan around them instead of pretending they are not there. Moving slowly carries its own cost, so we move fast and carefully at once.

The Safe Path and the Capable Path Are One.

We do not treat safety as a tax on capability. We engineer so the system that is safe to run is the same system that does the work, and we will hold a line we could cross rather than ship a capability we cannot stand behind. Safety is the engineering, not the brake.

Proven, Not Promised.

We publish what holds up: our research, our failures, and our open questions, and we pair every large claim with an honest hedge. We would rather show you a measured result than ask you to take a future on faith. The problems are bigger than any one organization, so we put the groundwork where others can build on it.

//THE MANDATE

The most important thing a powerful AI can do is not impress us, but be useful in the ways that let billions live longer, healthier, freer lives.

So we build toward that, on the path where the safe option and the capable option are the same, and we let ourselves be judged on results we can show you.

//PUBLICATIONS

The work, on the record.

Our team is researching and building now. Publications coming soon. We share what holds up, including the open problems we are still working through, so others can build on it.

Publication queue

How we measure real-world usefulnessComing soon

Safety evaluations before we shipComing soon

Reasoning under uncertaintyComing soon

Build the capability that earns its keep.

We are looking for researchers, engineers, and builders who want to work where frontier capability meets measurable human good, and who would rather prove an outcome than promise one.

See open roles Contact the team