@Matthewagi trained TERN-35B-A3B to try decision models. On an H100 PCIe it reached $0.0116/Mtok, then $0.00773/Mtok on the cost-score Pareto frontier.

Training note with a photo. @Matthewagi says the goal of the TypeSafe Jev model is to unlock Jevon's paradox. They trained TERN-35B-A3B to play with decision models, aiming at the price frontier. On an H100 PCIe they reached $0.0116 per million tokens; modifications reached the cost-score Pareto frontier at $0.00773 per million tokens. For people training their own decision models.