Revelo

Revelo is the only data firm at this scale focused exclusively on code-gen data. We run one of the largest networks of senior software engineers in the Americas (400k+), and that network provides the data infra that frontier labs use to improve their models' code-gen capabilities.

What's here

Our datasets follow one pattern: a free sample you can inspect row by row, and a gated extended set you can request access to.

• Samples are open (CC-BY) and drawn from the same production pipeline and annotator pool as the full sets. What you see is what you'd buy.
• Extended sets are gated. Request access with the form on the repo and we'll respond within 2 business days.

Current focus areas: hard SWE-Bench-style tasks with failure-annotated agent trajectories, Terminal-Bench 2.0-style tasks with step-level ATIF annotations, and design-quality data for AI-generated frontend code.

How the data gets made

Every dataset is expert-authored by senior engineers from our network, with multi-layer QA: peer review plus an audit pass by senior reviewers. Task difficulty is validated empirically through model rollouts, not guessed. Each dataset card documents the full methodology.

We also publish our own evals, starting with UI Quality Bench, a benchmark where 3 senior designers independently scored 350 AI-generated interfaces against a craft rubric.

Working with us

Custom collection, exclusivity, and volume licensing are all available. Start with any sample repo, then request access to an extended set, or go to revelo.com/ai to request data packs or talk to a researcher.