Exclusive: Nine-person Halluminate raises $30 million, counts four top U.S. AI labs as customers

Halluminate is trying to make AI better at finance by building the kind of practice rooms that general models often lack. The San Francisco startup, which says it has nine people, just raised $30 million in a Series A led by Oak HC/FT. That lifts its total funding to $38.5 million.

The company does two things: it benchmarks models on financial tasks to see where they fail, then turns those weak spots into simulated training environments. In one benchmark released in August, seven frontier models had to work through a fake acquisition due-diligence process. The test included 88 tasks based on anonymized private-equity transactions and was written and reviewed by practicing deal professionals. The top average score was 51%.

The details matter. One task pushed an agent to redline a statement of work using a 160-file data room, 21 emails across nine threads, and four meeting notes. As the terms changed, the system had to figure out which instructions were current and which were stale, while keeping the unchanged parts intact. The models often lost the thread. They missed required edits, used the wrong analysis, or leaned on information that had already been superseded.

That’s the problem Halluminate wants to sell against. CEO Jerry Wu thinks training data will become more specialized by industry, so a finance environment should look nothing like one for coding or health care. He calls the company’s systems “verticalized data research labs,” and he expects the complexity of those environments to keep rising as models improve. Internally, he says that every six to eight months the complexity has to roughly double just to stay useful.

Oak HC/FT’s Matt Streisfeld bought that pitch. He says finance is packed with long, messy knowledge work across banking, private equity, consulting, and accounting, and that the value of specialized testing will grow as AI agents take on work that stretches over hours or even days. Halluminate is keeping its focus narrow for now: frontier model labs first, enterprise customers later. According to Wu, four of the top five closed-source U.S. AI labs are already customers, and the company says it has crossed the mid-eight figures in annualized revenue run rate and is profitable.