Yachting Models is the internal control plane for YachtingStack's own fine-tuned models — the engines behind Agents and every deployment. It holds a canonical registry of every model and version, with explicit lineage from teacher to distilled student, so nothing is scattered across environment variables and people's heads.
A benchmark leaderboard scores every version on the same evaluation set — quality, latency and cost — so promoting a model into production is a decision made on evidence, not a guess. A synthetic-data and distillation pipeline is in development, turning large teacher models into smaller, cheaper student models without pooling private data into public ones.
Promotion into Agents runs through a manual, audited gate today, with a benchmark-gated automatic path planned next.