Model Judges Get Funded: Vals AI Said to Raise $40M for Evaluating Other AIs

Investor newsletters report that Vals AI, a U.S. startup focused on evaluating AI systems, is lining up a roughly $40 million Series A round led by Andreessen Horowitz, with participation from hedge‑fund‑backed investors and existing backers. While the company is still relatively small, the deal underscores growing demand for independent tools that can tell enterprises which models actually work best for real‑world tasks.

Vals AI is best known for its Finance Agent benchmark, which scores large language models on complex financial‑analysis workflows, and has become a reference point in recent academic work and model system cards from major labs. OpenAI, Anthropic and others now cite Finance Agent results when describing how their models perform on research, reasoning and reporting tasks in finance.

As generative AI moves from demos into production, companies face two questions: “Which model should we choose?” and “How do we know it’s safe and reliable?” Vals AI’s pitch is to automate what human experts would do—designing tasks, grading answers and comparing cost‑performance—so that dozens of models can be tested under consistent conditions. The anticipated funding round suggests that “AI that evaluates other AI” is becoming its own investment theme, and hints that model‑selection and governance tooling could be as strategically important as the models themselves.