Blog

We raised a $16M Series A led by Standard Capital to build the software that turns data into better AI. Reinforcement learning is how frontier models improve, and HUD is the platform where anyone can build and sell the RL environments that make that happen.

Our round included participation from Y Combinator, Exceptional Capital, Liquid 2 Ventures, and Twenty Two Ventures, with angel investments from Dylan Patel from SemiAnalysis, Roon from OpenAI, Ivan Burazin from Daytona, and Theo Browne.

• 1.3M+ — Task runs
• 2,500+ — Environments created
• 15M+ — Inference calls

How We Got Here

We started HUD after building computer use agents at Hume, where we kept hitting the same wall. The agents were improving fast, but the infrastructure to train and evaluate them barely existed, so we decided to build it.

We built our own benchmarks and they got used right away. Autonomy-10 evaluated OpenAI's Operator at launch. SheetBench-50 was validated by PwC, Cisco, Charles Schwab, and Fannie Mae. OSWorld-Verified fixed more than 300 issues from the original benchmark.

Our Mission

At HUD, we believe high quality data and evaluations will be critical to the development of agentic systems, ultimately leading to AGI. We designed the Autonomy evaluation as a first step for testing AI agents to navigate a diverse set of AI and browser tasks. As a basis for our work, we took principles from the ARC-AGI ^[1][2] and existing benchmarks ^{[3][4][5][6][7][8]} to create the most comprehensive eval framework for agentic AI models yet. Importantly, our benchmarks support computer, browser and tool use agents, and therefore test generalizable abilities relevant to everyday intelligence tasks.

Designing Autonomy-10

Autonomy-10 is only the first 10% of what we're building out for the most broad-scale AI agent benchmark. To maintain evaluation integrity and prevent overfitting, the full evaluation suite is kept private. We work directly with frontier labs and researchers to provide rigorous evaluations while contributing to open-source benchmarking efforts. Still, Autonomy-10 is already the most functionally diverse evaluation of the current state of the art models. We take inspiration from current agentic AI benchmarks ^[3] as well as the more general AI progress signals and safety evaluations via the Humanity's Last Exam program ^[10].