Recently
@Harvey introduced Tenet, its first model, trained for long horizon legal work.
Harvey post-trained it from a Kimi K3 base in collaboration with Fireworks using async RL on our Training API.
A thread on promising initial results for both performance and cost-efficiency🧵