Atlas

We are a team of developers, engineers, and data scientists who constantly found ourselves asking "What is the best AI model for X?". As it turns out, this was not a straightforward question. Most benchmarks for frontier AI models came from the model creators themselves, or were reliant on crowdsourced "arena" style leaderboards which often felt subjective. Objective benchmarks did exist, but there was no easy way to get independent results for them without setting up the pipelines yourself. We built Atlas by leveraging analytics as a base principle. It is our belief that generative AI should be held to the same standards as traditional software. Atlas currently has the largest, most extensive suite of benchmarks (over 50) out of any public leaderboard, and provides traces for individual prompts, which is something that no other leaderboard does.