How may we serve you?

we research novel architectural designs and develop open-weight models, testing different configurations to find what works best.

we're building an extensive training corpus for pre-training, fine-tuning, and general evaluation, plus new benchmarks to measure what improves.

we built our own inference stack to serve open-weight models privately. replicas across the U.S. keep latency low, and a tiered cache keeps long conversations cheap.

we measure our models on independent benchmarks, mostly ones built on private tests that nobody can train on, and weigh every score against how fast the model answers.

Cicero is our decision model, here it is against every decision model on OpenRouter, by full score on the Decision Index 0.4.1 and median latency.