Independent AI research lab
Intelligence
beyond scale.
We research new architectures and build infrastructure for more capable, efficient, and adaptive intelligent systems.
The research program
Can intelligence improve the way it improves?
We investigate architectures that can change how they reason—and experiments that establish whether those changes help.
Inside our researchWhat we’re working on
Research & engineeringArchitectures that adapt
Changing the structures of reasoning, while keeping evaluation independent of the system being changed.
Recursive self-improvementResearchReasoning that can be checked
Using Lean-verified theorem proving to study correctness, transfer, and the efficiency of reasoning.
Automated reasoningEngineering · In developmentRelay
An active Manaslu Labs project focused on efficient autonomous agent execution. Under development.
Discover RelayIntelligence should be scalable.
Compute alone isn’t the path.
Larger models are one route to more capable AI. We explore how better representations, reasoning mechanisms, and execution systems can open another.
Our long-term question is recursive: can improvements to a system’s problem-solving architecture help it discover the next improvement? Answering it requires more than a rising score. It requires transferable capability, retained knowledge, and complete accounting of the work.
Why Manaslu Labs existsAmbition, tested.
We make hypotheses explicit, compare changes under fixed budgets, and keep negative results in the record. Our work is experimental. Evidence decides what survives.
How we evaluate progressAn open conversation