mendotaAI
The frontier

The next
frontier

Intelligence begins in the real world. We build the foundations to bring it there.

Research

Research,
out in the world.

Real environments. Verifiable progress. Our benchmarks put agents in front of real infrastructure and real incidents, beyond the sandbox.

InfraBench

Real infrastructure. Real incidents. A benchmark for AI agents beyond the sandbox.

152 incidents·4 layers·3 backends

Explore InfraBench

BenchScope

How many independent signals does your benchmark provide? A diagnostic for measurement breadth, built on effective dimensionality, that flags the redundant parts of a suite.

22 benchmarks·8 domains·8,400+ evaluations

Read BenchScope
The lab

The next frontier
is what AI can do.

Mendota AI explores the space between model capability and real-world consequence. Our starting point: rigorous benchmarks that make that gap visible.

Based in San Francisco, we're building toward intelligence grounded in the world's experience.

Our core researchers come from Harvard, MIT, and the University of Washington, and include core contributors to Terminal-Bench.

See open roles contact@mendota.ai
Grounded

Intelligence,
grounded.

Built from the world's experience. Proven in the work.

Explore our research