Vals is a startup that wants to change how AI companies test their models. It was founded in 2024 by a small group that included 25-year-old Rayan Krishnan.
The company just closed a $40 million Series A funding round. Andreessen Horowitz led the round, which closed last month.
Vals had already raised a seed round before this. That earlier round was led by 8VC and Bloomberg Beta.
Krishnan previously interned at Palantir. He studied at Stanford and worked with Microsoft and the school's AI lab as an undergraduate.
How Vals Tests AI Models
Krishnan says older benchmarking systems could not keep up with newer AI models. He wanted a system that measured real world skills instead of general knowledge.
Many benchmarking systems publish their test questions online. This means companies can train their models on the answers ahead of time.
Vals does not release its test materials to the public. The company says this keeps its results more accurate.
Vals tests models on tasks tied to specific fields. These include law, finance, and coding work.
The company also checks for negative outcomes, not just positive ones. It studies what could go wrong if a model made mistakes in the real world.
Vals has added new testing areas over time. These now include cybersecurity, biosecurity, mental health, and recursive self improvement.
Growth And What Comes Next
Vals makes money by charging companies to test their AI models. Krishnan compares this to a student paying a fee to take the SAT.
The company's revenue is now eight times higher than it was last year. Its staff has grown just as fast.
Vals started the year with eight employees. That number has since tripled to 25 people.
Krishnan said the company plans to move into a larger office soon. He also wants to hire between 10 and 15 more people.
Vals recently launched a program that provides model evaluations to federal agencies. This adds government work to its list of clients.
Krishnan expects benchmarking to matter more as AI companies go public. He pointed to SpaceX's public listing and Anthropic's planned listing later this year.
He said he expects OpenAI to go public soon as well. Krishnan believes evaluations like his will help these companies explain their AI investments to the public.
