Vals Raises $40M Series A for Private, Task-Based AI Model Evaluations
The AMW Read
The Series A is an incremental but meaningful financing update for a specialized evaluation provider serving a growing AI testing market.
Vals Raises $40M Series A for Private, Task-Based AI Model Evaluations
Vals, founded in 2024, has raised a $40 million Series A led by Andreessen Horowitz, following a seed round led by 8VC and Bloomberg Beta. The company builds undisclosed evaluations intended to test whether AI models can complete real-world legal, finance, and coding tasks, rather than optimizing for public academic benchmarks. It also assesses potential harms spanning cybersecurity, biosecurity, mental health, and recursive self-improvement. Vals says revenue grew eightfold year over year and its team reached 25 employees.
The funding underscores a growing market for evaluation infrastructure as model developers, enterprise buyers, and public-sector users need evidence that systems work beyond benchmark leaderboards. Keeping test material private is central to Vals' proposition: public benchmarks can become targets for model training, weakening their value as an independent signal. Its federal-agency evaluation program also broadens the buyer set from model labs toward institutions that need repeatable, defensible testing before deployment.
For builders, the practical implication is that task-level evaluation must become part of the product-development loop, not a final marketing exercise. For investors, Vals' model will depend on whether private benchmarks become a recurring operational expense for developers and enterprises, and whether its test design remains credible as models and deployment risks change quickly.
