Recent Releases
View moreThe Pathway to Infinite Intelligence
Historically, the level of complexity in the simulations used for training has guided model capabilities. Combine that with scaling laws, and our conviction in infinite intelligence reduces to researching, building, and scaling the most complex simulated environments to let frontier AI labs train more capable and safer models.
- 1Infra
- 2Data
- 3Environments
- 4Rollouts
- 5Grading
Optimizing for Research Throughput
We believe great research is fundamentally an infrastructure problem. Testing every variation of an idea, collecting data at scale, and iterating on robust grading are all bottlenecked by the number of rollouts you can run. So we build our post-training stack to optimize for throughput, our guiding light in scaling simulation complexity.
Get in touch
Frontier Labs and Researchers
Talk to our team about how we can help you in your RL, research, and evaluations.
Contact usJoin us
We are hiring exceptional talent across distributed systems, evaluation research, and post-training.
About us