Vals wants to become tech's benchmark authority for AI models

Vals, a startup founded in 2024, is trying to fix what it sees as a broken system for evaluating AI models. In under two years, the company has built a notable presence in the industry: last year it closed a seed round led by 8VC and Bloomberg Beta, and last month it raised $40 million in a Series A led by Andreessen Horowitz.

Co-founder Rayan Krishnan, 25, previously interned at Palantir and worked at Microsoft and Stanford’s artificial intelligence lab as an undergraduate. He says Vals was born from watching benchmarking fall behind the pace of AI progress. “We were seeing a bunch of new, very capable models come to market quickly, and the academic benchmarks [were] not keeping up with that frontier advance,” Krishnan said.

Unlike many benchmarking systems that publish their test materials — which lets companies train against the tests — Vals keeps its evaluations private. Rather than measuring general knowledge, it tests models on complex, industry-specific tasks in fields like law, finance, and coding, aiming to check whether a model can produce work of the same quality as a human, and to flag negative outcomes as well as positive ones.

The company’s evaluation scope keeps expanding. Krishnan said Vals is building benchmarks on recursive self-improvement and doing work in mental health, cybersecurity, biosecurity, and how models apply the Geneva Convention under the law of armed conflict. It also recently launched a program offering model evaluations to federal agencies.

Companies pay Vals to test their own models — a business model Krishnan compares to a student paying the College Board to take the SAT — because the results help them find and fix weaknesses. Those evaluations are also becoming a factor companies weigh when deciding whether to acquire new AI models.

Vals says its revenue is now eight times what it was a year ago. Its headcount has also grown fast: the company started the year with eight people and has already tripled to a team of 25, with Krishnan planning to hire another 10 to 15 people and move to a bigger office.

Krishnan expects benchmarking to become more central as AI companies face public markets — pointing to SpaceX’s IPO, Anthropic’s, which he said is slated for later this year, and his suspicion that OpenAI will go public soon.

Why it matters

As AI models become embedded across industries and their makers face public-market scrutiny, independent benchmarks like Vals’ could become a standard reference point for how companies justify their technology’s value to investors and regulators.