Development process and initial evaluation of a 13.1 billion parameter AI model released

Riemann Computing released the development process and initial evaluation results for "My Name is Riemann," a roughly 13.1-billion-parameter language model built by progressively scaling up from a 163-million-parameter OpenPeerLLM base rather than training a same-sized model from scratch. The technical documentation was published October 28 through the OpenPeer AI organization, with the company detailing training methods, hardware, and early benchmarks in a December 4 statement. The staged approach matters because it offers a lower-cost path to large models by reusing existing open weights, though the disclosed evaluation is preliminary and the model is a pre-release version. What to watch next is fuller benchmark results and whether the scaling recipe holds at larger sizes.