We build small (≤500M parameter) task-specific models for finance under Apache 2.0 — engineered for sub-second CPU inference, public eval tables, and drop-in compatibility with the pipelines quant teams actually run.
