Halluminate raises $30 million as 4 top US AI labs pay

Halluminate was founded in 2024 and is betting that the data and environments used to train AI will become increasingly specialized by industry. CEO Jerry Wu frames the company's systems as "verticalized data research labs," arguing that simulating an investment banker's work is fundamentally different from simulating a software engineer's.

The company's approach starts with benchmarks that expose where AI models fall short on financial tasks, then turns those failure modes into reinforcement-learning environments. Its August benchmark, built from anonymized private-equity transactions and reviewed by practicing deal professionals, found that the best of seven frontier models averaged just 51% across 88 tasks — agents left out required changes, used the wrong analytical method, or relied on superseded information.

Oak HC/FT General Partner Matt Streisfeld said finance's breadth of complex knowledge work and Halluminate's expertise stood out, and that testing and specialization will matter more as agents take on long-horizon work. Wu's "Moore's law of environments" — his estimate that complexity must roughly double every six to eight months — suggests the company's ability to keep pace with frontier labs, rather than the size of this round, is what will determine whether its early customer concentration pays off. For now, enterprise customers and other industries are explicitly not a priority.