Frontier labs
Research collaboration, not just a vendor. We co-design environments, evals, and data around your training priorities, and bring the domain judgment to know which capabilities are worth climbing toward.
An applied AI research lab
Research partners on the climb to safe, aligned intelligence.
We work side by side with frontier labs and enterprises to build the environments, evaluations, and data that move models from capable to expert.
Build with usThe name
culmen (Latin, n.): the summit; the highest point.
Models won’t reach their potential alone. We join research teams as partners on the climb, bringing domain expertise, research talent, and hard-to-get data to every stage, from first eval to trained model.
Who we partner with
Enterprises
Evaluate and improve models on your own workflows. We help you measure what matters, find failure modes before deployment, and train toward the work your teams actually do.
What we build
RL environments
Realistic, verifiable environments and tasks that push models to the edge of what they can do.
Evaluations & benchmarks
Rigorous evals built by domain experts, so you know where models fail before your users do.
Expert data
Rubrics, reasoning traces, and demonstrations from practitioners who do the work every day.
Data partnerships
Access to proprietary longitudinal and multimodal data, sourced and structured for training.
Research direction
Domain judgment on what to train for, so compute goes to capabilities that matter in the real world.
Domains
- Medicine & clinical care
- Life sciences & biology
- Software & code
- Robotics
- And beyond
We are researchers from the frontier, coming together to develop safe, aligned, and intelligent models across domains.