Harvey Tenet
Harvey's first in-house legal model, built by further training an open-weight Kimi K3.
Harvey Tenet is the first model that Harvey trained itself rather than renting from a big lab. It was not built from scratch. Harvey started from a version of Kimi K3, the open-weight model published by the Chinese lab Moonshot AI, and then trained it further on mock disputes and case files until it behaved like a lawyer rather than a generalist.
Harvey says Tenet performs at frontier level on well-known legal benchmarks while costing far less to run than the largest general models. That claim has not been independently verified. The interesting part is the recipe: take a strong open model, add deep domain data, and you get a specialist without a billion dollar training run.