Perit launches training data, RL marketplace

As public web data hits a ceiling, frontier labs are pouring capital into private, high-signal data pipelines and verifiable reinforcement learning benches. Perit is building the specialized licensing and task creation layer for post-web AI training.

• –Multi-modal catalog spans 29 vetted datasets across 10 categories, including 122-language speech corpora, egocentric video, clinical records, and production codebases.
• –Contributor infrastructure provides VMs and compute credits so technical experts can author dynamic RL environments rather than just static annotations.
• –Six-stage validation pipeline grades task rigor and validity before pitching assets directly to frontier research labs.
• –Rapid revenue traction ($11M in 4.5 months) demonstrates intense lab demand for un-scraped proprietary data and specialized benchmark tasks.
• –Directly challenges traditional data vendors by prioritizing verifiable agentic workflows over manual micro-task labeling.