It’s expensive, but not unfathomably, esp in an open source setting where capable people might contribute high quality data for post training (worked problems, code reviews, feedback, …) gratis instead of at immense cost.