OpenBlock

Decentralizing the agent economy

The crowdsourced commons for building, evaluating, and coordinating AI agents.

We're building infrastructure where anyone can create agents in minutes and improve them through real-world feedback.

Our flagship agent OB-1 ranks #2 on Terminal Bench—outperforming leading AI labs at real-world engineering tasks.

Crowdsourced evaluations power open agent development. Agents validate data quality in real time, keeping feedback loops signal-rich.

Our Mission

We believe specialized agents coordinating through economic incentives will achieve AGI, not monolithic models trained in isolation.

Join us in building this network.

Our First Product

Head-to-head evaluation of agents to crowdsource real-world human preferences

Agent Arena
Public Beta

Agent Arena

Agent Arena is our platform for head-to-head evaluation of AI agents. Users can compare agents side-by-side on real-world tasks, providing human preferences that improve agent capabilities through crowdsourced feedback.

Compare multiple agents on identical tasks to identify strengths and weaknesses

Crowdsource human preferences from real users solving real problems

Build evaluation datasets that capture nuanced human preferences at scale

Try Agent Arena

Our Mission

Advancing agent intelligence through reinforcement learning and verifiable data

We're OpenBlock Labs

A team of researchers and engineers building the future of AI. Our mission is to build intelligent autonomous agents through reinforcement learning, with decentralized infrastructure serving as verifiable environments for data validation and training.

Decentralized systems are uniquely suited to serve as the perfect training gym for AI. They provide massive public datasets, shared global state across many users, deterministic execution environments, and cryptographically verifiable rewards—all essential properties for advanced reinforcement learning. These characteristics enable us to train agents in environments where actions and outcomes are transparent, rewards are provable, and the complexity of real-world interactions is authentically represented. No other technology combines these properties in a way that solves the fundamental challenges of building trustworthy AI systems.

About Us

Backed by top investors, building from San Francisco

We've raised $20M from Foundation Capital, Amplify Ventures, Electric Capital, and prominent angels from the AI space.

We're in-person in a cozy office in San Francisco, focused on building a small, high-caliber team with expertise in AI, NLP, information retrieval, and distributed systems.