Datapoint AI

Datapoint AI provides human preference data infrastructure for teams developing and evaluating multimodal models. Its API streams people’s judgments to support online reinforcement learning, model refinement, and benchmarking, with the company advertising more than 10,000 preferences per minute across 200-plus countries. Teams can request comparisons, ratings, rankings, multiple-choice responses, or text answers for image, video, chat, and audio tasks.

The service is designed to bring human feedback into different stages of model development, from training loops and checkpoint comparisons to ongoing evaluations after launch. Datapoint AI also supports targeted studies focused on particular abilities, such as instruction following or tone, and recruitment by factors including locale, language, age, or profession. The company says responses pass checks for identity, behavior, attention, and bias, with embedded attention checks intended to screen out inattentive participants.

Additional offerings include API documentation, an SDK for online reinforcement learning, research grants, and DP-Bench, a leaderboard based on eligible human preference votes. Teams can contact Datapoint AI to discuss their task requirements and expected data volume.