DatologyAI launches Curation Studio data platform on October 10 for customizable frontier data curation
DatologyAI launched Curation Studio on October 10, automating a four-stage Clean, Calibrate, Create, Compose pipeline for frontier model training data. Across 39 public benchmarks, a 30B-A3B model reached 46.8% vs 37.7% baseline, while a 12B-A1.4B model matched performance using one-fifth the pretraining compute.
DatologyAI launched Curation Studio on October 10, a customizable data platform for frontier model training. The platform breaks training data preparation into four stages: Clean, Calibrate, Create, and Compose, packaging scattered curation work into a reusable engineering pipeline. Across 39 public benchmarks, a 30B-A3B model trained with the pipeline reached 46.8% versus the 37.7% baseline, a 9 percentage-point gain attributable to data quality and distribution calibration. A 12B-A1.4B model trained on one-fifth of the pretraining compute matched the 30B baseline, demonstrating that the pipeline can replicate larger-model results on smaller ones. Curation Studio runs in a customer private VPC, which allows enterprise teams to integrate proprietary training corpora without exposing raw data to external services. DatologyAI, founded in 2023, focuses on frontier data curation and works with compute partners including Together AI and several frontier model labs.